Michael Taylor
Contents
0.
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
A.
B.
C.
D.
E.
F.
G.
H.
I.
J.
K.
Poincare metrics
The fundamental theorem of algebra (elementary proof)
The Weierstrass approximation theorem
Inner product spaces
2 is irrational
Eulers constant
Rapid evaluation of the Weierstrass function
Introduction
This text covers material presented in complex analysis courses I have taught numerous
times at UNC. The core idea of complex analysis is that all the basic functions that arise
in calculus, first derived as functions of a real variable, such as powers and fractional
powers, exponentials and logs, trigonometric functions and their inverses, and also a host
of more sophisticated functions, are actually naturally defined for complex arguments, and
are complexdifferentiable (a.k.a. holomorphic). Furthermore, the study of these functions
on the complex plane reveals their structure more truly and deeply than one could imagine
by only thinking of them as defined for real arguments.
An introductory 0 defines the algebraic operations on complex numbers, say z = x + iy
and w = u + iv, discusses the magnitude z of z, defines convergence of infinite sequences
and series, and derives some basic facts about power series
(i.1)
f (z) =
ak z k ,
k=0
such as the fact that if this converges for z = z0 , then it converges absolutelyPfor z < R =
z0 , to a continuous function. It is also shown that, for z = t real, f 0 (t) = k1 kak tk1 ,
for R < t < R. Here we allow ak C. We define the exponential function
(i.2)
X
1 k
e =
z ,
k!
z
k=0
d at
e = aeat .
dt
We use this differential equation to derive further properties of the exponential function.
While 0 develops calculus for complex valued functions of a real variable, 1 introduces
calculus for complex valued functions of a complex variable. We define the notion of complex differentiability. Given an open set C, we say a function f : C is holomorphic
on provided it is complex differentiable, with derivative f 0 (z), and f 0 is continuous on .
Writing f (z) = u(z) + iv(z), we discuss the CauchyRiemann equations for u and v. We
also introduce the path integral and provide some versions of the fundamental theorem of
calculus in the complex setting. (More definitive results will be given in 5.)
In 2 we return to convergent power series and show they produce holomorphic functions.
We extend results of 0 from functions of a real variable to functions of a complex variable.
Section 3 returns to the exponential function ez , defined above. We extend (i.3) to
(i.4)
d az
e = aeaz .
dz
We show that t 7 et maps R onetoone and onto (0, ), and define the logarithm on
(0, ), as its inverse:
(i.5)
x = et t = log x.
We also examine the behavior of (t) = eit , for t R, showing that this is a unitspeed
curve tracing out the unit circle. From this we deduce Eulers formula,
(i.6)
z a = ea log z ,
a C, z C \ (, 0],
and investigate their basic properties. We also discuss inverse trigonometric functions in
the complex plane.
In 5 we introduce a major theoretical tool of complex analysis, the Cauchy integral
theorem. We provide a couple of proofs, one using Greens theorem and one based simply
on the chain rule and the fundamental theorem of calculus. Cauchys integral theorem
leads to Cauchys integral formula, and then to the general development of holomorphic
functions on a domain C in power series about any p , convergent in any disk
centered at p and contained in .
Results of 5 are applied in 6 to prove a maximum principle for holomorphic functions,
and also a result called Liouvilles theorem, stating that a holomorphic function on C that
is bounded must be constant. We show that each of these results imply the fundamental
theorem of algebra, that every nonconstant polynomial p(z) must vanish somewhere in C.
In 7 we discuss harmonic functions on planar regions, and their relationship to holomorphic functions.
In 8 we establish Moreras theorem, a sort of converse to Cauchys integral theorem.
We use this in 9 to prove Goursats theorem, to the effect that the C 1 hypothesis can be
dropped in the characterization of holomorphic functions.
After a study of the zeros and isolated singularities of holomorphic functions in 10
11, we look at other infinite series developments of functions: Laurent series in 12 and
Fourier series in 13. The method we use to prove a Fourier inversion formula also produces
a Poisson integral formula for the unique harmonic function in a disk with given boundary
values. Variants of Fourier series include the Fourier transform and the Laplace transform,
discussed in 1415. These transforms provide some interesting examples of integrals
whose evaluations cannot be done with the techniques of elementary calculus.
Residue calculus, studied in 16, provides a powerful tool for the evaluation of many definite integrals. A related tool with many important applications is the argument principle,
studied in 17. In a sense these two sections lie at the heart of the course.
In 18 and 19 we make use of many of the techniques developed up to this point to
study two special functions, the Gamma function and the Riemann zeta function. In both
cases these functions are initially defined in a halfplane and then analytically continued
as meromorphic functions on C.
Sections 2028 have a much more geometrical flavor than the preceding sections. We
look upon holomorphic diffeomorphisms as conformal maps. We are interested in holomorphic covering maps and in implications of their existence. We also study the Riemann
sphere, as a conformal compactification of the complex plane. We include two gems of
nineteenth century analysis, the Riemann mapping theorem and Picards theorem. An
important tool is the theory of normal families, studied in 21.
In 29 we return to a more functiontheoretic point of view. We establish Harnack
estimates for harmonic functions and use them to obtain Liouville theorems of a more
general nature than obtained in 7. These tools help us produce some concrete illustrations
of Picards theorem, such as the fact that ez z takes on each complex value infinitely
often.
In 3033 we provide an introduction to doubly periodic meromorphic functions, also
known as elliptic functions, and their connection to theta functions and elliptic integrals,
and in 34 we show how constructions of compact Riemann surfaces provide tools in elliptic
function theory.
This text concludes with several appendices. In Appendix A we collect material on
metric spaces and compactness, including particularly the ArzelaAscoli theorem, which is
an important ingredient in the theory of normal families. We also prove the contraction
mapping theorem, of use in Appendix B. In Appendix B we discuss the derivative of
a function of several real variables and prove the Inverse Function Theorem, in the real
context, which is used in 4 to get the Inverse Function Theorem for holomorphic functions
on domains in C.
Appendix C discusses metric tensors on surfaces. It is of use in 26, on the Riemann
sphere and other Riemann surfaces. This material is also useful for Appendix E, which
introduces a special metric tensor, called the Poincare metric, on the unit disk and other
domains in C, and discusses some connections with complex function theory, including
another proof of Picards big theorem. In between, Appendix D proves Greens theorem
for planar domains, of use in one proof of the Cauchy integral theorem in 5.
In Appendix F we give a proof of the Fundamental Theorem of Algebra that is somewhat
different from that given in 7. It is elementary, in the sense that it does not rely on
results from integral calculus.
In Appendix G we show that a construction arising in 14 to prove the Fourier inversion
formula also helps establish a classical result of Weierstrass on approximating continuous
functions by polynomials. In fact, this proof is basically that of Weierstrass, and I prefer
it to other proofs that manage to avoid complex function theory.
Appendix H deals with inner product spaces, and presents some results of use in our
treatments of Fourier series and the Fourier transform, in 1314. In Appendix I, we
present a proof that 2 is irrational. Appendix J has material on Eulers constant. Appendix K complements 32 with a description of how to evaluate (z) rapidly via the use
of theta functions.
Acknowledgment
Thanks to Shrawan Kumar for testing this text in his Complex Analysis course, for pointing
out corrections, and for other valuable advice.
z = x + iy,
where x and y are real numbers. These numbers make up the complex plane, which is
just the xyplane with the real line forming the horizontal axis and the real multiples of i
forming the vertical axis. See Figure 0.1. We write
(0.2)
x = Re z,
y = Im z.
(0.3)
(0.4)
z + w = w + z,
zw = wz,
z + (w + c) = (z + w) + c,
z(wc) = (zw)c,
c(z + w) = cz + cw,
as following from their counterparts for real numbers. If c 6= 0, we can perform division
by c,
z
= w z = wc.
c
See (0.13) for a neat formula.
For z = x+iy, we define z to be the distance of z from the origin 0, via the Pythagorean
theorem:
p
(0.8)
z = x2 + y 2 .
Note that
(0.9)
z2 = z z,
where z = x iy,
z + z = 2 Re z,
z z = 2i Im z,
and
(0.11)
z + w = z + w,
zw = z w.
(0.13)
The following result is known as the triangle inequality, as Figure 0.2 suggests.
Proposition 0.1. Given z, w C,
(0.14)
z + w z + w.
= zz + ww + zw + wz
= z2 + w2 + 2 Re(zw),
while
(0.16)
Thus (0.14) follows from the inequality Re(zw) zw, which in turn is immediate from
the definition (0.8). (For any C, Re .)
We can define convergence of a sequence (zn ) in C as follows. We say
(0.17)
10
zn z, wn w = zn + wn z + w and zn wn zw,
zn zm  0 as m, n .
As in (0.18), this holds if and only if (xn ) and (yn ) are Cauchy in R. The following is an
important fact.
(0.19B)
A detailed presentation of the field R of real numbers, including a proof of (0.19C), is given
in Chapter 1 of [T0].
We can define the notion of convergence of an infinite series
(0.20)
zk
k=0
sn =
n
X
zk .
k=0
sn w =
zk = w.
k=0
Note that
n+m
sn+m sn  =
zk
k=n+1
(0.23)
n+m
X
k=n+1
zk .
11
(0.24)
zk  < ,
k=0
(0.24A)
zk  A,
N.
k=0
Then the sequence (sn ) given by (0.21) is Cauchy, hence the series (0.20) is convergent.
Proof. If (sn ) is not Cauchy, there exist a > 0, n % , and m > 0 such that
sn +m sn  a.
Passing to a subsequence, one can assume that n+1 > n + m . Then (0.23) implies
mX
+n
zk  a,
k=0
contradicting (0.24A).
If (0.24) holds, we say the series (0.20) is absolutely convergent.
An important class of infinite series is the class of power series
(0.25)
ak z k ,
k=0
with ak C. Note that if z1 6= 0 and (0.25) converges for z = z1 , then there exists C <
such that
ak z1k  C,
(0.25A)
k.
X
k=0
ak z  C
X
k=0
rk =
C
< ,
1r
the last identity being the classical geometric series computation. This yields the following.
12
Proposition 0.2. If (0.25) converges for some z1 6= 0, then either this series is absolutely
convergent for all z C, or there is some R (0, ) such that the series is absolutely
convergent for z < R and divergent for z > R.
We call R the radius of convergence of (0.25). In case of convergence for all z, we say
the radius of convergence is infinite. If R > 0 and (0.25) converges for z < R, it defines
a function
(0.27)
f (z) =
ak z k ,
z DR ,
k=0
zn z = f (zn ) f (z).
Proof. For each z DR , there exists S < R such that z DS , so it suffices to show that
f is continuous on DS whenever 0 < S < R. Pick T such that S < T < R. We know that
there exists C < such that ak T k  C for all k. Hence
z DS = ak z k  C
(0.30)
S k
T
SN (z) =
N
X
ak z k ,
RN (z) =
k=0
ak z k .
k=N +1
Each SN (z) is a polynomial in z, and it follows readily from (0.19) that SN is continuous.
Meanwhile,
(0.32)
z DS RN (z)
X
k=N +1
k
X
S
ak z  C
= CN ,
T
k
k=N +1
13
N 0, as N .
Then f is continuous on DS .
Proof. Let zn z in DS . We need to show that, given > 0, there exists M = M () <
such that
f (z) f (zn ) , n M.
To get this, pick N such that (0.32A) holds with N = /3. Now use continuity of SN , to
deduce that there exists M such that
SN (z) SN (zn )
,
3
n M.
f (t) =
a k tk
k=0
converges for real t satisfying t < R. Then f is differentiable on the interval R < t < R,
and
(0.34)
f (t) =
kak tk1 ,
k=1
14
We first check absolute convergence of the series (0.34). Let S < T < R. Convergence
of (0.33) implies there exists C < such that
ak T k C,
(0.35)
k.
Hence, if t S,
kak tk1 
(0.36)
C S k
k
,
S
T
g(t) =
kak tk1
k=1
is continuous on (R, R). To show that f 0 (t) = g(t), by the fundamental theorem of
calculus, it is equivalent to show
Z
(0.38)
g(t) =
bk tk
k=0
converges for real t, satisfying t < R. Then, for t < R,
Z
(0.40)
0
X
bk k+1
g(s) ds =
t
,
k+1
k=0
k k+1
t
Rbk tk ,
k+1
(0.41)
SN (t) =
N
X
k=0
bk t ,
RN (t) =
X
k=N +1
bk tk .
15
so
Z
(0.44)
0
Z t
N
X
bk k+1
g(s) ds =
t
+
RN (s) ds,
k+1
0
k=0
and
(0.45)
Z t
Z t
RN (s) ds
RN (s) ds CRN .
f 0 (t) = f (t),
f (0) = 1.
We look for a solution as a power series, of the form (0.33). If there is a solution of this
form, (0.34) requires
(0.47)
a0 = 1,
ak+1 =
ak
,
k+1
X
1 k
f (t) = e =
t ,
k!
t
t R.
k=0
This defines the exponential function et . Convergence for all t follows from the ratio test.
(Cf. Exercise 4 below.) More generally, we define
(0.49)
X
1 k
e =
z ,
k!
z
z C.
k=0
Again the ratio test shows that this series is absolutely convergent for all z C. Another
application of Proposition 0.4 shows that
(0.50)
eat =
X
ak
k=0
k!
tk
solves
(0.51)
d at
e = aeat ,
dt
16
whenever a C.
We claim that eat is the only solution to
f 0 (t) = af (t),
(0.52)
f (0) = 1.
d at
e f (t) = aeat f (t) + eat af (t) = 0,
dt
(0.53)
where we use the product rule, (0.51) (with a replaced by a), and (0.52). Thus eat f (t)
is independent of t. Evaluating at t = 0 gives
eat f (t) = 1,
(0.54)
t R,
whenever f (t) solves (0.52). Since eat solves (0.52), we have eat eat = 1, hence
eat =
(0.55)
1
,
eat
t R, a C.
Thus multiplying both sides of (0.54) by eat gives the asserted uniqueness:
f (t) = eat ,
(0.56)
t R.
(0.57)
d at bt (a+b)t
e e e
dt
= aeat ebt e(a+b)t beat ebt e(a+b)t + (a + b)eat ebt e(a+b)t
= 0,
t R.
t R, a, b C,
or, setting t = 1,
(0.60)
ea+b = ea eb ,
a, b C.
17
We will resume study of the exponential function in 3, and derive further important
properties.
Exercises
1. Supplement (0.19) with the following result. Assume there exists A > 0 such that
zn  A for all n. Then
(0.61)
2. Letting sn =
zn z =
Pn
k=0
(1 r)sn = 1 rn+1 ,
(0.62)
1
1
.
zn
z
hence sn =
1 rn+1
.
1r
Deduce that
(0.63)
0 < r < 1 = sn
1
, as n ,
1r
as stated in (0.26).
3. The absolute convergence said to follow from (0.36) can be stated as follows:
(0.64)
0 < r < 1 =
k=1
Prove this.
Hint. Writing r = s2 , 0 < s < 1, deduce (0.64) from the assertion
(0.65)
a > 0 =
k
is bounded, for k N.
(1 + a)k
Show that
(0.67)
(1 + a)k = (1 + a) (1 + a) 1 + ka,
a > 0, k N.
18
4. This exercise discusses the ratio test, mentioned in connection with the infinite series
(0.49). Consider the infinite series
(0.69)
ak ,
ak C.
k=0
k+1
(0.70)
k N =
r.
ak
Show that
(0.71)
ak  < .
k=0
ak  aN 
k=N
r` =
`=0
aN 
.
1r
5. In case
(0.73)
ak =
zk
,
k!
show that for each z C, there exists N < such that (0.70) holds, with r = 1/2.
6. This exercise discusses the integral test for absolute convergence of an infinite series,
which goes as follows. Let f be a positive, monotonically decreasing, continuous function
on [0, ), and suppose ak  = f (k). Then
Z
ak  <
N
X
k=1
f (t) dt < .
0
k=0
Prove this.
Hint. Use
Z
ak 
f (t) dt
0
N
1
X
k=0
X
1
< a > 1.
a
n
n=1
ak .
19
8. This exercise deals with alternating series. Assume bk & 0. Show that
(1)k bk is convergent,
k=0
n+m
(1)k bk bn .
k=n
9. Show that
k=1 (1)
d
f (t)g(t) = f 0 (t)g(t) + f (t)g 0 (t).
dt
(0.74)
(0.75)
k = 1, 2, 3, . . . ,
hence
Z
(0.76)
0
sk ds =
1 k+1
t
,
k+1
k = 0, 1, 2, . . . .
Note the use of these identities in (0.44), leading to many of the identities in (0.34)(0.51).
20
lim
h0
1
[f (z + h) f (z)] = a.
h
lim
h1 0
1
f
[f (z + h1 ) f (z)] =
(z),
h1
x
and
(1.3)
lim
h2 0
1
1 f
[f (z + ih2 ) f (z)] =
(z),
ih2
i y
In the first case, the limit exists and we have f 0 (z) = 1 for all z. In the second case, the
limit does not exist. The function f (z) = z is not complexdifferentiable.
Definition. A function f : C is holomorphic if and only if it is complexdifferentiable
and f 0 is continuous on .
Adding the hypothesis that f 0 is continuous makes for a convenient presentation of the
basic results. In 9 it will be shown that every complex differentiable function has this
additional property.
So far, we have seen that f1 (z) = z is holomorphic. We produce more examples of
holomorphic functions. For starters, we claim that fk (z) = z k is holomorphic on C for
each k Z+ , and
(1.5)
d k
z = kz k1 .
dz
21
d
(f g)(z) = f 0 (z)g(z) + f (z)g 0 (z).
dz
(1.7)
1
[f (z + h)g(z + h) f (z)g(z)]
h
1
= [f (z + h)g(z + h) f (z)g(z + h) + f (z)g(z + h) f (z)g(z)]
h
1
1
= [f (z + h) f (z)]g(z + h) + f (z) [g(z + h) g(z)].
h
h
The first term in the last line tends to f 0 (z)g(z), and the second term tends to f (z)g 0 (z),
as h 0. This gives (1.6). If f 0 and g 0 are continuous, the right side of (1.6) is also
continuous, so f g is holomorphic.
It is even easier to see that the sum of two holomorphic functions is holomorphic, and
(1.8)
d
(f (z) + g(z)) = f 0 (z) + g 0 (z),
dz
d 1
1
= 2.
dz z
z
In fact,
(1.10)
1h 1
1i
1
h
1
=
=
,
h z+h z
h z(z + h)
z(z + h)
X
1 k
e =
z ,
k!
z
k=0
d z
e = ez .
dz
22
(1.13)
Hence
(1.14)
1 z+h
eh 1
[e
ez ] = ez
.
h
h
k=1
k=0
eh 1 X 1 k1 X
1
=
h
=
hk ,
h
k!
(k + 1)!
and hence
(1.16)
eh 1
= 1.
lim
h0
h
This gives (1.12). Another proof of (1.12) will follow from the results of 2.
We next establish a chain rule for holomorphic functions. In preparation for this,
we note that the definition (1.1) of complex differentiability is equivalent to the condition
that, for h sufficiently small,
(1.17)
with
(1.18)
lim
h0
r(z, h)
= 0,
h
r(z, h) = o(h).
f g(z) = f (g(z)),
is holomorphic, and
(1.20)
d
f (g(z)) = f 0 (g(z))g 0 (z).
dz
23
This implies f g is complexdifferentiable and gives (1.20). Since the right side of (1.20)
is continuous, f g is seen to be holomorphic.
Combining Proposition 1.2 with (1.9), we have the following.
Proposition 1.3. If f : C is holomorphic, then 1/f is holomorphic on \ S, where
(1.24)
S = {z : f (z) = 0},
and, on \ S,
(1.25)
d 1
f 0 (z)
=
.
dz f (z)
f (z)2
d f (z)
e
= f 0 (z)ef (z) .
dz
f
f
and
exist, and are continuous on ,
x
y
and
(1.28)
1 f
f
=
x
i y
24
When (1.27) holds, one says f is of class C 1 and writes f C 1 (). As shown in
Appendix B, if f C 1 (), it is Rdifferentiable, i.e.,
(1.29)
A=
f
(z),
x
B=
f
(z).
y
A=
1
B = a,
i
leading back to (1.28). This gives the following converse to Proposition 1.4.
Proposition 1.5. If f : C is C 1 and (1.28) holds, then f is holomorphic.
The equation (1.28) is called the CauchyRiemann equation. Here is an alternative
presentation. Write
(1.33)
u = Re f, v = Im f.
u
v
=
,
x
y
v
u
= .
x
y
To pursue this a little further, we change perspective, and regard f as a map from an
open subset of R2 into R2 . We represent an element of R2 as a column vector. Objects
on C and on R2 correspond as follows.
On C
z = x + iy
(1.35)
f = u + iv
h = h1 + ih2
On R2
x
z=
y
u
f=
v
h1
h=
h2
25
R(z, h) = o(h).
(1.37)
Df (z) =
u/x
v/x
u/y
v/y
u
v
, =
(1.38)
Df =
, =
.
x
x
Now the map z 7 iz is a linear transformation on C R2 , whose 2 2 matrix representation is given by
0 1
(1.39)
J=
.
1 0
Note that, if L =
, then
(1.40)
JL =
, LJ =
,
so JL = LJ if and only if = and = . (When JL = LJ, we say J and L commute.)
When L = Df (z), this gives (1.38), proving the following.
Proposition 1.6. If f C 1 (), then f is holomorphic if and only if, for each z ,
(1.41)
In the calculus of functions of a real variable, the interaction of derivatives and integrals,
via the fundamental theorem of calculus, plays a central role. We recall the statement.
Theorem 1.7. If f C 1 ([a, b]), then
Z
(1.42)
d
dt
g(s) ds = g(t).
a
26
In the study of holomorphic functions on an open set C, the partner of d/dz is the
integral over a curve, which we now discuss.
A C 1 curve (or path) in is a C 1 map
: [a, b] ,
where [a, b] = {t R : a t b}. If f : C is continuous, we define
Z
(1.44)
f (z) dz =
the right side being the standard integral of a continuous function, as studied in beginning
calculus (except that here the integrand is complex valued). More generally, if f, g : C
are continuous and = 1 + i2 , with j real valued, we set
Z
(1.44A)
f (z) dx + g(z) dy =
a
(1.46)
The proof of Proposition 1.9 is essentially the same as that of Proposition 1.2. To
address Proposition 1.8, we have
Z
Z
0
f (z) dz =
a
(1.47)
f 0 ((t)) 0 (t) dt
d
f ((t)) dt
a dt
= f ((b)) f ((a)),
=
27
the second identity by (1.46) and the third by (1.42). This gives Proposition 1.8.
The second half of Theorem 1.7 involves producing an antiderivative of a given function
g. In the complex context, we have the following.
Definition. A holomorphic function g : C is said to have an antiderivative f on
provided f : C is holomorphic and f 0 = g.
Calculations done above show that g has an antiderivative f in the following cases:
(1.48)
g(z) = z k ,
g(z) = ez ,
1
z k+1 , k 6= 1,
k+1
f (z) = ez .
f (z) =
g(z) dz = 0
for any closed path in , i.e., any C 1 path : [a, b] such that (a) = (b). In 3,
we will see that if is the unit circle centered at the origin,
Z
(1.50)
1
dz = 2i,
z
(1.51)
f (z) = i
g(a + is) ds +
b
28
(1.53)
We also have
(1.54)
f
(z) = ig(a + iy) +
y
Z
a
g
(t + iy) dt,
y
(1.55)
Z x
g
1 f
= g(a + iy) +
(t + iy) dt
i y
a t
= g(a + iy) + [g(x + iy) g(a + iy)]
= g(z).
Comparing (1.54) and (1.55), we have the CauchyRiemann equations for f , and Proposition 1.10 follows.
Examples of open sets satisfying (1.51) include disks and rectangles, while C \ 0 does
not satisfy (1.51), as one can see by taking a + ib = 1, x + iy = 1.
Exercises
1. Let f, g C 1 (), not necessarily holomorphic. Show that
(1.56)
1
gx (z)
=
,
x g(z)
g(z)2
1
gy (z)
=
.
y g(z)
g(z)2
f (z)
f (z)
and
.
x g(z)
y g(z)
29
(c)
z+1
,
z2 + 1
z+1
f (z) = 2
,
z +1
f (z) = e1/z ,
(d)
f (z) = ez .
(a)
f (z) =
(b)
(a)
f (z) =
(b)
f (z) = zez ,
(c)
f (z) = z 2 + ez ,
(a)
f (z) = z,
(b)
f (z) = z,
(c)
f (z) =
1
,
(z + 5)2
f (z) = ez .
(d)
6. Do Exercise 5 with
(t) = t4 + it2 .
R
7. Recall the definition (1.44) for f (z) dz when f C() and : [a, b] is a C 1
curve. Suppose s : [a, b] [, ] is C 1 , with C 1 inverse, such that s(a) = , s(b) = . Set
(s(t)) = (t). Show that
Z
Z
f (z) dz =
f () d,
30
1
(f (z + h) f (z)).
h
Hint. Write
h (f g)(z) = h f (z) g(z + h) + f (z)h g(z).
10. Show that, for each > 0, A < ,
h z n nz (n+1)
uniformly on {z C : < z A}.
Hint. Use induction.
31
f (z) =
an (z z0 )n .
n=0
Recall from 0 that to such a series there is associated a radius of convergence R [0, ],
with the property that the series converges absolutely whenever z z0  < R (if R > 0),
and diverges whenever z z0  > R (if R < ). We begin this section by identifying R as
follows:
1
(2.2)
= lim sup an 1/n .
R
n
This is established in the following result, which reviews and complements Propositions
0.20.3.
Proposition 2.1. The series (2.1) converges whenever z z0  < R and diverges whenever
z z0  > R, where R is given by (2.2). If R > 0, the series converges uniformly on
{z : z z0  R0 }, for each R0 < R. Thus, when R > 0, the series (2.1) defines a
continuous function
(2.3)
f : DR (z0 ) C,
where
(2.4)
0
(2.5)
z z0  < R0 < R = an (z z0 )n 
,
0
R
for n N , so (2.1) is dominated by a convergent geometrical series in DR0 (z0 ).
For the converse, we argue as follows. Suppose R00 > R, so infinitely many an 1/n
1/R00 , hence infinitely many an (R00 )n 1. Then
z z n
0
00
n
z z0  R > R = infinitely many an (z z0 ) 
1,
00
R
forcing divergence for z z0  > R.
The assertions about uniform convergence and continuity follow as in Proposition 0.3.
The following result, which extends Proposition 0.4 from the real to the complex domain,
is central to the study of holomorphic functions. A converse will be established in 5, as a
consequence of the Cauchy integral formula.
32
Proposition 2.2. If R > 0, the function defined by (2.1) is holomorphic on DR (z0 ), with
derivative given by
0
(2.6)
f (z) =
nan (z z0 )n1 .
n=1
Proof. Absolute convergence of (2.6) on DR (z0 ) follows as in the proof of Proposition 0.4.
Alternatively (cf. Exercise 3 below), we have
(2.7)
lim n1/n = 1 = lim sup nan 1/n = lim sup an 1/n ,
so the power series on the right side of (2.6) converges locally uniformly on DR (z0 ), defining
a continuous function g : DR (z0 ) C. It remains to show that f 0 (z) = g(z).
To see this, consider
(2.8)
fk (z) =
k
X
an (z z0 ) ,
gk (z) =
n=0
k
X
nan (z z0 )n1 .
n=1
We have fk f and gk g locally uniformly on DR (z0 ). By (1.12) we have fk0 (z) = gk (z).
Hence it follows from Proposition 1.8 that, for z DR (z0 ),
Z
(2.9)
fk (z) = a0 + gk () d,
z
for any path z : [a, b] DR (z0 ) such that z (a) = z0 and z (b) = z. Making use of the
locally uniform convergence, we can pass to the limit in (2.9), to get
Z
(2.10)
f (z) = a0 + g() d.
z
x0
and hence
f
(z) = g(z),
x
(2.11)
y0
33
and hence
f
(z) = ig(z).
y
(2.12)
A=
n ,
B=
n=0
n ,
n=0
AB =
n ,
n =
n=0
Proof. Take Ak =
Pk
n=0 n , Bk =
(2.15)
n
X
j nj .
j=0
Pk
n=0
n . Then
Ak Bk =
k
X
n + Rk
n=0
with
(2.16)
Rk =
m n ,
(m,n)(k)
Hence
Rk 
(2.17)
mk/2 k/2nk
n  + B
nk/2
m  n  +
X
m  n 
k/2mk nk
m ,
mk/2
where
(2.18)
A=
X
n=0
n  < ,
B=
n  < .
n=0
It follows thatPRk 0 as k . Thus the left side of (2.15) converges to AB and the
right side to n=0 n . The absolute convergence of (2.14) follows by applying the same
argument with n replaced by n  and n replaced by n .
34
Corollary 2.4. Suppose the following power series converge for z < R:
X
X
(2.19)
f (z) =
an z n , g(z) =
bn z n .
n=0
n=0
f (z)g(z) =
cn z ,
cn =
n=0
n
X
aj bnj .
j=0
The following result, which is related to Proposition 2.3, has a similar proof.
P
P
Proposition 2.5. If ajk C and j,k ajk  < , then j ajk is absolutely convergent
P
for each k, k ajk is absolutely convergent for each j, and
XX
X
X X
(2.21)
ajk =
ajk =
ajk .
j
j,k
Proof. Clearly the hypothesis implies j ajk  < for each k and
j. It also implies that there exists B < such that
SN =
N X
N
X
ajk  B,
P
k
N.
j=0 k=0
Now, whenever M, K N ,
M X
K
N X
N
X
ajk
ajk
j=0 k=0
so
j=0 k=0
X
(j,k)C(N )
M X
N X
N
X
ajk
ajk
j=0 k=0
and hence
j=0 k=0
j=0 k=0
ajk ,
(j,k)C(N )
N X
N
X
ajk
ajk
j=0 k=0
ajk ,
ajk .
(j,k)C(N )
We have a similar result with the roles of j and k reversed, and clearly the two finite sums
agree. It follows that
X
X
X
ajk
ajk < 2, > 0,
j=0 k=0
k=0 j=0
yielding (3.23).
Using Proposition 2.5, we demonstrate the following.
35
Proposition 2.6. If (2.1) has a radius of convergence R > 0, and z1 DR (z0 ), then f (z)
has a convergent power series about z1 :
(2.22)
f (z) =
bk (z z1 )k ,
k=0
The proof of Proposition 2.6 will not use Proposition 2.2, and we can use this result
to obtain a second proof of Proposition 2.2. Shrawan Kumar showed the author this
argument.
an ( + z1 )n
n=0
X
n
X
n=0 k=0
n k nk
an
z1 ,
k
the second identity by the binomial formula (cf. (2.34) below). Now,
(2.24)
X
n
X
n=0 k=0
X
n
k
nk
an 
 z1 
=
an ( + z1 )n < ,
k
n=0
provided  + z1  < R, which is the hypothesis in (2.22) (with z0 = 0). Hence Proposition
2.5 gives
(2.25)
fz1 () =
X
k=0 n=k
n nk k
an
z
.
k 1
bk =
X
n=k
n nk
an
z
.
k 1
b1 =
X
n=1
nan z1n1 .
36
Second proof of Proposition 2.2. The result (2.22) implies f is complex differentiable
at each z1 DR (z0 ), and the computation (2.27) translates to (2.6), with z = z1 .
Remark. The result of Proposition 2.6 is a special case of a general result on representing
a holomorphic function by a convergent power series, which will be established in 5.
Exercises
1. Determine the radius of convergence R for each of the following series. If 0 < R < ,
examine when convergence holds at points on z = R.
(a)
zn
n=0
(b)
X
zn
n
n=1
(c)
X
zn
n2
n=1
(d)
X
zn
n!
n=0
(e)
X
zn
2n
n=0
(f)
X
z2
2n
n=0
2. Show that if the power series (2.1) has radius of convergence R > 0, then f 00 , f 000 , . . .
are holomorphic on DR (z0 ) and
(2.28)
f (n) (z0 ) = n! an .
Here we set f (n) (z) = f 0 (z) for n = 1, and inductively f (n+1) (z) = (d/dz)f (n) (z).
37
(2.29)
(2.30)
and hence
lim n1/n = 1,
(2.31)
(2.32)
n
X
n k
(1 + a) =
a ,
k
n
k=0
n
n!
.
=
k!(n k)!
k
Another version is
(2.34)
n
X
n k nk
(z + w) =
z w
.
k
n
k=0
Verify this identity and show that (2.33) implies (2.29) when a > 0.
Hint. To verify (2.34), expand
(2.35)
(z + w)n = (z + w) (z + w)
as a sum of monomials and count the number of terms equal to z k wnk . Use the fact that
n
(2.36)
= number of combinations of n objects, taken k at a time.
k
5. As a special case of Exercise 2, note that, given a polynomial
(2.37)
p(z) = an z n + + a1 z + a0 ,
38
we have
p(k) (0) = k! ak ,
(2.38)
0 k n.
Apply this to
pn (z) = (1 + z)n .
(2.39)
(k)
Compute pn (z), using (1.5), then compute p(k) (0), and use this to give another proof of
(2.33), i.e.,
(2.40)
n
X
n k
pn (z) =
z ,
k
k=0
n!
n
.
=
k!(n k)!
k
39
(3.1)
X zj
z2
zj
e =1+z+
+ +
+ =
.
2!
j!
j!
j=0
z
By the ratio test this converges for all z, to a continuous function on C. Furthermore, the
exponential function is holomorphic on C. This function satisfies
d z
e = ez ,
dz
(3.2)
e0 = 1.
One derivation of this was given in 1. Alternatively, (3.2) can be established by differentiating term by term the series (3.1) to get (by Proposition 2.2)
(3.3)
X 1
d z X
1
e =
z j1 =
zk
dz
(j 1)!
k!
j=1
k=0
X
k=0
1 k
z = ez .
k!
dj z
e = ez ,
j
dz
j = 1, 2, 3, . . . .
By (2.21), any function f (z) that is the sum of a convergent power series about z = 0 has
the form
(3.5)
f (z) =
X
f (j) (0)
j=0
j!
zj ,
which for a function satisfying (3.2) and (3.4) leads to (3.1). A simple extension of (3.2) is
(3.6)
d az
e = a eaz .
dz
ez ew = ez+w ,
z, w C.
40
For an alternative proof, we can expand the left side of (3.7) into a double series:
ez ew =
(3.8)
X
X
z j X wk
z j wk
=
.
j!
k!
j!k!
j=0
k=0
j,k=0
z+w
X
(z + w)n
=
,
n!
n=0
n
X
n j nj
(z + w) =
z w
,
j
j=0
n
n
n!
.
=
j!(n j)!
j
Setting k = n j, we have
(3.11)
z+w
n=0 j+k=n;j,k0
X
z j wk
1 n! j k
z w =
.
n! j!k!
j!k!
j,k=0
See (2.14) for the last identity. Comparing (3.8) and (3.11) again gives the identity (3.7).
We next record some properties of exp(t) = et for real t. The power series (3.1) clearly
gives et > 0 for t 0. Since et = 1/et , we see that et > 0 for all t R. Since
det /dt = et > 0, the function is monotone increasing in t, and since d2 et /dt2 = et > 0,
this function is convex. Note that
(3.12)
et > 1 + t,
for t > 0.
Hence
(3.13)
lim et = +.
t+
Since et = 1/et ,
(3.14)
lim et = 0.
As a consequence,
(3.15)
exp : R (0, )
is smooth and onetoone and onto, with positive derivative, so the inverse function theorem
of onevariable calculus applies. There is a smooth inverse
(3.16)
L : (0, ) R.
41
log x = L(x).
(3.18)
gives
L0 (et )et = 1,
(3.19)
hence L0 (et ) =
1
,
et
i.e.,
d
1
log x = .
dx
x
(3.20)
Since log 1 = 0, we get
Z
(3.21)
log x =
1
dy
.
y
x, y (0, ),
log : C \ R C,
d
1
log z = ,
dz
z
log 1 = 0.
log z =
1
d
,
the integral taken along any path in C\R from 1 to z. Comparison with (3.21) shows that
this function restricted to (0, ) coincides with log as defined in (3.17). In 4 we display
log in (3.22A) as the inverse of the exponential function exp(z) = ez on the domain
= {x + iy : x R, y (, )}, making use of some results that will be derived next.
(See also Exercises 1315 at the end of this section.)
42
(3.23)
t R.
This traces out a curve in the complex plane, and we want to understand which curve it
is. Let us set
eit = c(t) + is(t),
(3.24)
with c(t) and s(t) real valued. First we calculate eit 2 = c(t)2 + s(t)2 . For x, y R,
(3.25)
z = x + iy = z = x iy = zz = x2 + y 2 = z2 .
It is elementary that
(3.26)
z, w C = zw = z w = z n = z n ,
and z + w = z + w.
Hence
(3.27)
ez
X
zk
k=0
k!
= ez .
In particular,
(3.28)
Hence t 7 (t) = eit has image in the unit circle centered at the origin in C. Also
(3.29)
(0) = 1,
0 (0) = i.
Thus, for t between 0 and the circumference of the unit circle, the arc from (0) to (t) is
an arc on the unit circle, pictured in Figure 3.3, of length
Z
(3.31)
`(t) =
 0 (s) ds = t.
Standard definitions from trigonometry say that the line segments from 0 to 1 and from
0 to (t) meet at angle whose measurement in radians is equal to the length of the arc of
the unit circle from 1 to (t), i.e., to `(t). The cosine of this angle is defined to be the
43
xcoordinate of (t) and the sine of the angle is defined to be the ycoordinate of (t).
Hence the computation (3.31) gives
(3.32)
c(t) = cos t,
s(t) = sin t.
(3.33)
(3.34)
applied to (3.33), yields
d
cos t = sin t,
dt
(3.35)
d
sin t = cos t.
dt
We can use (1.3.7) to derive formulas for sin and cos of the sum of two angles. Indeed,
comparing
ei(s+t) = cos(s + t) + i sin(s + t)
(3.36)
with
(3.37)
gives
(3.38)
Derivations of the formulas (3.35) for the derivative of cos t and sin t given in first
semester calculus courses typically make use of (3.38) and further limiting arguments,
which we do not need with the approach used here.
The standard definition of the number is half the length of the unit circle. Hence
is the smallest positive number such that (2) = 1. We also have
(3.39)
() = 1,
= i.
1
3
= +
i,
2
2
3 1
+ i.
2
2
44
p
(s) = ( 1 s2 , s),
0 s = sin .
`=
 (s) ds =
0
ds
.
1 s2
(3.43)
0
ds
= ,
1 s2
sin = ,
when 0 < < /2. As anotherway to see this, note that the substitution s = sin gives,
by (3.35), ds = cos d, while 1 s2 = cos by (3.28), which implies
cos2 t + sin2 t = 1.
(3.44)
Thus
Z
(3.45)
0
ds
=
1 s2
d = ,
0
=
6
1/2
dx
.
1 x2
One can produce a power series for (1 y)1/2 and substitute y = x2 . (For more on
this, see the exercises at the end of 5.) Integrating the resulting series term by term, one
obtains
(3.47)
an 1 2n+1
=
,
6
2n
+
1
2
n=0
a0 = 1,
an+1 =
2n + 1
an .
2n + 2
45
Using a calculator, one can sum this series over 0 n 20 and show that
(3.49)
= 3.141592653589 .
Exercises
Heres another way to demonstrate the formula (3.35) for the derivatives of sin t and cos t.
1. Suppose you define cos t and sin t so that (t) = (cos t, sin t) is a unitspeed parametrization of the unit circle centered at the origin, satisfying (0) = (1, 0), 0 (0) = (0, 1), (as we
did in (3.32)). Show directly (without using (3.35)) that
0 (t) = ( sin t, cos t),
and hence deduce (3.35). (Hint. Differentiate (t) (t) = 1 to deduce that, for each
t, 0 (t) (t). Meanwhile, (t) =  0 (t) = 1.)
2. It follows from (3.33) and its companion eit = cos t i sin t that
(3.50)
cos z =
eiz + eiz
,
2
sin z =
eiz eiz
2i
cos z =
X
(1)k
k=0
(2k)!
2k
z ,
X
(1)k
sin z =
z 2k+1 .
(2k + 1)!
k=0
3. Extend the identities (3.35) and (3.38) to complex arguments. In particular, for z C,
we have
(3.52)
cos(z + 2 ) = sin z,
sin(z +
2)
= cos z,
cos(z + ) = cos z,
cos(z + 2) = cos z,
sin(z + ) = sin z,
sin(z + 2) = sin z.
4. We define
(3.53)
cosh y =
ey + ey
,
2
sinh y =
ey ey
.
2
46
tan z =
sin z
,
cos z
cot z =
cos z
.
sin z
Show that
(3.56)
tan(z + 2 ) = cot z,
tan(z + ) = tan z,
and
(3.57)
d
1
tan z =
= 1 + tan2 z.
dz
cos2 z
6. For each of the following functions g(z), find a holomorphic function f (z) such that
f 0 (z) = g(z).
a) g(z) = z k ez , k Z+ .
b) g(z) = eaz cos bz.
7. Concerning the identities in (3.40), verify algebraically that
3 3
+
i = 1.
2
2
Then use ei/6 = ei/2 ei/3 to deduce the stated identity for (/6) = ei/6 .
8. Let be the unit circle centered at the origin in C, going counterclockwise. Show that
Z
1
dz =
z
ieit
dt = 2i,
eit
as stated in (1.50).
9. One sets
sec z =
1
,
cos z
so (3.57) yields (d/dz) tan z = sec2 z. Where is sec z holomorphic? Show that
d
sec z = sec z tan z.
dz
10. Show that
1 + tan2 z = sec2 z.
47
and
d
(sec z + tan z) = sec z (sec z + tan z),
dz
d
(sec z tan z) = sec z tan2 z + sec3 z
dz
= 2 sec3 z sec z.
= {x + iy : x R, y (, )}.
13. Using ex+iy = ex eiy and the properties of exp : R (0, ) and of (y) = eiy
established in (3.15) and (3.33), show that
(3.59)
log(ez ) = z
z .
Hint. Apply the chain rule to compute g 0 (z) for g(z) = log(ez ). Note that g(0) = 0.
15. Deduce from Exercises 13 and 14 that
log : C \ R
is onetoone and onto, and is the inverse to exp in (3.59).
48
Dg(y) = Df (x)1 .
A proof of this is given in Appendix B. This result has the following consequence, which
is the Inverse Function Theorem for holomorphic functions.
Theorem 4.2. Let C be open and let f : C be holomorphic. Take p
and assume f 0 (p) 6= 0. Then there exists a neighborhood O of p and a neighborhood U of
q = f (p) such that f : O U is onetoone and onto, the inverse g = f 1 : U O is
holomorphic, and, for z O, w = f (z),
(4.2)
g 0 (w) =
1
f 0 (z)
Proof. If we check that g is holomorphic, then (4.2) follows from the chain rule, Proposition
1.2, applied to
g(f (z)) = z.
We know g is C 1 . By Proposition 1.6, g is holomorphic on U if and only if, for each w U ,
Dg(w) commutes with J, given by (1.39). Also Proposition 1.6 implies Df (z) commutes
with J. To finish, we need merely remark that if A is an invertible 2 2 matrix,
(4.3)
AJ = JA A1 J = JA1 .
As a first example, consider the function Sq(z) = z 2 . Note that we can use polar
coordinates, (x, y) = (r cos , r sin ), or equivalently z = rei , obtaining z 2 = r2 e2i . This
shows that Sq maps the right halfplane
(4.4)
H = {z C : Re z > 0}
Sqrt : C \ R H,
49
given by
(4.6)
We also write
z 1/2 = Sqrt(z).
(4.7)
We can define other nonintegral powers of z on C \ R . Before doing so, we take a look
at log, the inverse function to the exponential function, exp(z) = ez . Consider the strip
(4.8)
= {x + iy : x R, < y < }.
(4.9)
Note that dez /dz = ez is nowhere vanishing, so (4.9) has a holomorphic inverse we denote
log:
log : C \ R .
(4.10)
Note that
(4.11)
log 1 = 0.
where the integral is taken along any path from 1 to z in C \ R . Comparison with (3.22B)
shows that the function log produced here coincides with the function arising in (3.22A).
(This result also follows from Exercises 1315 of 3.)
Now, given any a C, we can define
(4.14)
z a = Powa (z),
Powa : C \ R C
by
z a = ea log z .
(4.15)
The identity eu+v = eu ev then gives
(4.16)
z a+b = z a z b ,
a, b C, z C \ R .
In particular, for n Z, n 6= 0,
(4.17)
(z 1/n )n = z.
Making use of (4.12) and the chain rule (1.20), we see that
d a
(4.18)
z = a z a1 .
dz
While Theorem 4.1 and Corollary 4.2 are local in nature, the following result can provide
global inverses, in some important cases.
50
f (z1 ) f (z0 )
(4.19)
a
=
af 0 (1 t)z0 + tz1 dt,
z1 z0
0
which has positive real part and hence is not zero.
As an example, consider the strip
o
n
e = x + iy : < x < , y R .
(4.20)
2
2
0
Take f (z) = sin z, so f (z) = cos z. It follows from (3.54) that
(4.21)
e
Re cos z = cos x cosh y > 0, for z ,
1
1
,
2i
e under z 7 eiz is the right half plane H, given by (4.4). Below we will
and the image of
show that the image of H under g is
(4.23)
e
z .
g 0 (z) = (1 z 2 )1/2 ,
sin
z=
(1 2 )1/2 d,
where the integral is taken along any path from 0 to z in C\{(, 1][1, )}. Compare
this identity with (3.43), which treats the special case of real z (1, 1).
It remains to prove the asserted mapping property of g, given in (4.22). We rephrase
the result for
1
1
(4.28)
h() = g(i) =
+
.
2
51
Proposition 4.4. The function h given by (4.28) maps both the upper half plane U = { :
Im > 0} and the lower half plane U = { : Im < 0}, onetoone onto the set (4.23).
Proof. Note that h : C \ 0 C, and
(4.29)
= h().
2 2w + 1 = 0,
with solutions
(4.31)
=w
p
w2 1.
Thus, for each w C, there are two solutions, except for w = 1, with single solutions
h(1) = 1, h(1) = 1. If we examine h(x) for x R \ 0 (see Fig. 4.1), we see that
w (, 1] [1, ) if and only if R. If w belongs to the set (4.23), h() = w has
two solutions, both in C \ R, and by (4.29) they are reciprocals of each other. Now
(4.32)
= 2,

Exercises
1. Show that, for z < 1,
(4.33)
log(1 + z) =
(1)n1
n=1
Z
log(1 + z) =
0
zn
.
n
1
d,
1+
X
(1)n1
= log 2.
n
n=1
52
Hint. Using properties of alternating series, show that, for r (0, 1),
(4.34A)
N
X
(1)n1 n
r = log(1 + r) + N (r),
n
n=1
N (r)
rN +1
.
N +1
log x
= 0.
x
x+
1 z 2 + iz,
z = sin w,
sin1 z =
1
log 1 z 2 + iz ,
i
z < 1.
Hint. Consult Fig. 4.2. To get the last step, it helps to show that
R(z) =
1z
1+z
53
tan
z=
0
d
,
1 + 2
where tan1 is the inverse of tan in Exercise 7, and the integral is over any path from 0
to z within the region (4.36).
9. Show that
d
log sec z = tan z.
dz
and sec3 z.
1 + t2 dt,
on an interval [0, x] R.
(a) Evaluate (4.38) using the change of variable t = tan and the results of Exercise 10.
(b) Evaluate (4.38) using the change of variable t = sinh u and the results of Exercise 12
in 3.
12. As a variant of (4.20)(4.23), show that z 7 sin z maps
n
o
54
satisfying (4.27), in a manner that avoids appeal to Theorem 4.2. (Compare the construction of log as an inverse to exp in Exercises 1315 of 3.) We will use the result
that
(4.40)
G0 (z) = (1 z 2 )1/2 ,
G : C,
G(0) = 0.
G(z) =
(1 2 )1/2 d,
z ,
G(sin z) = z,
e
z .
Hint. Apply the chain rule to f (z) = G(sin z), making use of (4.25), to show that f 0 (z) = 1
e
for all z .
16. Use (4.40) and Exercise 15 to show that
(4.44)
55
17. Expanding on Proposition 4.4, show that the function h, given by (4.28), has the
following properties:
(a) h : C \ 0 C is onto,
(b) h : C \ {0, 1, 1} C \ {1, 1} is twotoone and onto,
(c) h : R \ 0 R \ (1, 1) is onto, and is twotoone, except at x = 1.
18. Given a C \ R , set
Ea (z) = az = ez log a ,
z C.
x2 + y 2 = =
hence
r
x=
+a
,
2
2xy = b,
p
a2 + b2 ,
y=
b
.
2x
56
ZZ
Z
f
g
dx dy = (f dx + g dy).
x y
Z
f dz =
f (dx + i dy).
ZZ
f
f
f dz =
i
dx dy.
x
y
Whenever f is holomorphic, the integrand on the right side of (5.3) vanishes, so we have
the following result, known as Cauchys integral theorem:
Theorem 5.2. If f C 1 () is holomorphic, then
Z
(5.4)
f (z) dz = 0.
1
f (z0 ) =
2i
f (z)
dz.
z z0
57
Proof. Note that g(z) = f (z)/(z z0 ) is holomorphic on \ {z0 }. Let Dr be the open disk
of radius r centered at z0 . Pick r so small that Dr . Then (5.4) implies
Z
Z
f (z)
f (z)
(5.6)
dz =
dz.
z z0
z z0
Dr
To evaluate the integral on the right, parametrize the curve Dr by () = z0 +rei . Hence
dz = irei d, so the integral on the right is equal to
Z
(5.7)
0
f (z0 + rei )
irei d = i
rei
f (z0 + rei ) d.
1
f (z0 ) =
2
2
0
1
f (z0 + re ) d =
`(Dr )
f (z) ds(z),
Dr
(n)
n!
(z) =
2i
f ()
d.
( z)n+1
1
(f (z + h) f (z)),
h
1
h f (z) =
2i
Z
h
1
f () d.
z
58
Now, as h 0, given z
/ ,
h ( z)1 ( z)2 , uniformly on ,
by Exercise 8 of 1, and this gives (5.10). The formula (5.11) follows inductively, using
Z
(n 1)!
1
(n1)
h f
(z) =
h
f () d,
2i
( z)n
(5.12)
1
1 X z z0 n
=
,
( z0 ) (z z0 )
z0 n=0 z0
f (z) =
an (z z0 )n ,
n=0
with
(5.17)
1
an =
2i
f ()
f (n) (z0 )
d
=
.
( z0 )n+1
n!
Remark. The second identity in (5.17) follows from (5.11). Alternatively, once we have
(5.16), it also follows from (2.28) that an is equal to the last quantity in (5.17).
Next we use the Cauchy integral theorem to produce an integral formula for the inverse
of a holomorphic map.
59
is independent of s.
To set things up, say s (t) = (s, t) is C 2 for a s b, t R, and periodic of period 1
in t. Denote the left side of (5.20) by (s):
Z 1
(5.21)
(s) =
f (s, t) t (s, t) dt.
0
Hence
Z
(5.22)
(s) =
0
f (s, t) s (s, t) dt
0 t
(5.23)
Z 1
0
=
f (s, t) t (s, t) s (s, t) + f (s, t) t s (s, t) dt.
0
60
f 0 ()s t = f 0 ()t s ,
we see that the right sides of (5.22) and (5.23) are equal. But the fundamental theorem of
calculus implies the left side of (5.23) is equal to
(5.25)
f (s, 1) s (s, 1) f (s, 0) s (s, 0) = 0.
Thus 0 (s) = 0 for all s, and the proposition is proven.
For a variant of Proposition 5.7, see Exercise 14.
Our third proof of Cauchys theorem establishes a result slightly weaker than Theorem
5.1, namely the following.
Proposition 5.8. With as in Theorem 5.1, assume O, open in C, and f is
holomorphic on O. Then (5.4) holds.
The proof will not use Greens thorem. Instead, it is based on Propositions 1.8 and
1.10.
By Proposition 1.8, if f had a holomorphic antiderivative on O, then wed have
R
f
(z)
dz = 0 for each closed path in O. Since is a union of such closed paths,
this would give (5.4). As we have seen, f might not have an antiderivative on O, though
Proposition 1.10 does give conditions guaranteeing the existence of an antiderivative. The
next strategy is to chop some neighborhood of in O into sets to which Proposition 1.10
applies.
To carry this out, tile the plane C with closed squares Rjk of equal size, with sides
parallel to the coordinate axes, and check whether the following property holds:
(5.26)
See Fig. 5.1 for an example of part of such a tiling. If (5.26) fails, produce a new tiling by
dividing each square into four equal subsquares, and check (5.26) again. Eventually, for
example when the squares have diameters less than dist(, O), which is positive, (5.26)
must hold.
If Rjk 6= , denote this intersection by jk . We have = j,k jk , and furthermore
Z
X Z
(5.27)
f (z) dz =
f (z) dz,
j,k
jk
the integrals over those parts of jk not in cancelling out. Now Proposition 1.10
implies f has a holomorphic antiderivative on each Rjk O, and then Proposition 1.8
implies
Z
(5.28)
f (z) dz = 0,
jk
so (5.4) follows.
Finally, we relax the hypotheses on f , for a certain class of domains . To specify the
class, we say an open set O C is star shaped if there exists p O such that
(5.28A)
0 < a < 1, p + z O = p + az O.
61
Theorem 5.9. Let be a bounded domain with piecewise smooth boundary. Assume can
be partitioned into a finite number of piecewise smoothly bounded domains j , 1 j K,
such that each j is star shaped. Assume f C() and that f is holomorphic on . Then
(5.4) holds.
Proof. Since
Z
(5.28B)
f (z) dz =
K Z
X
f (z) dz,
j=1
j
it suffices to prove the result when itself is star shaped, so (5.28A) holds with O = ,
p . Given f C(), holomorphic on , define
(5.28C)
fa : C,
fa (p + z) = f (p + az),
0 < a < 1.
Exercises
1. Using (5.10), show that if fk are holomorphic on C and fk f locally uniformly,
then f is holomorphic and fk f locally uniformly.
2. Show that, for z < 1, C,
(5.29)
(1 + z) =
an ()z n ,
n=0
an () =
( 1) ( n + 1)
.
n!
(1 z ) =
(1)n an ()z 2n ,
n=0
62
with an () as above. Take = 1/2 and verify that (3.46) yields (3.47).
4. Suppose f is holomorphic on a disk centered at 0 and satisfies
f 0 (z) = af (z),
for some a C. Prove that f (z) = Keaz for some K C.
Hint. Find the coefficients in the power series for f about 0. Alternative. Apply d/dz to
eaz f (z).
5. Suppose f : C C is a function that is not identically zero. Assume that f is
complexdifferentiable at the origin, with f 0 (0) = a. Assume that
f (z + w) = f (z)f (w)
for all z, w C. Prove that f (z) = eaz .
Hint. Begin by showing that f is complexdifferentiable on all of C.
6. Suppose f : C is holomorphic, p , and f (p) = 0. Show that g(z) = f (z)/(z p),
defined at p as g(p) = f 0 (p), is holomorphic. More generally, if f (p) = = f (k1) (p) = 0
and f (k) (p) 6= 0, show that f (z) = (z p)k g(z) with g holomorphic on and g(p) 6= 0.
Hint. Consider the power series of f about p.
7. For f as in Exercise 6, show that on some neighborhood of p we can write f (z) =
[(z p)h(z)]k , for some nonvanishing holomorphic function h.
8. Suppose f : C is holomorphic and onetoone. Show that f 0 (p) 6= 0 for all p .
Hint. If f 0 (p) = 0, then apply Exercise 7 (to f (z) f (p)), with some k 2. Apply
Theorem 4.2 to the function G(z) = (z p)h(z).
Reconsider this problem when you get to 11, and again in 17.
9. Assume f : C is holomorphic, Dr (z0 ) , and f (z) M for z Dr (z0 ). Show
that
(5.32)
f (n) (z0 )
M
n.
n!
r
63
f (z) dz = 0.
11. Take = {z C : 0 < z < 2}, and let (t) = eit , 0 t 2. Calculate
deduce that is not simply connnected.
dz/z and
is independent of s.
For more on this, see Exercise 8 of 7.
64
(6.1)
f (z0 + rei ) d,
valid whenever
(6.2)
Dr (z0 ) = {z C : z z0  r} ,
ZZ
f (z) dx dy =
(6.3)
Dr (z0 )
f (z0 + sei )s ds d
= r2 f (z0 ),
or
(6.4)
1
f (z0 ) =
Ar
ZZ
f (z) dx dy,
Dr (z0 )
where Ar = r2 is the area of the disk Dr (z0 ). This is another form of the mean value
property. We use it to prove the following result, known as the maximum principle for
holomorphic functions.
Proposition 6.1. Let C be a connected, open set. If f is holomorphic on , then,
given z0 ,
(6.5)
Proof. In the latter context, f  must assume a maximum at some point in (cf. Proposition A.14). Hence it suffices to prove (6.5).
65
Thus, assume there exists z0 such that the hypotheses of (6.5) hold. Set
(6.7)
O = { : f () = f (z0 )}.
f (0 ) = f (z),
z D (0 ).
f (z) z.
Furthermore, equality holds in (6.10), for some z D1 (0) \ 0, if and only if f (z) = cz for
some constant c of absolute value 1.
Proof. The hypotheses imply that g(z) = f (z)/z is holomorphic on D1 (0) (cf. 5, Exercise
6), and that g(z) 1/a on the circle {z : z = a}, for each a (0, 1). Hence the
maximum principle implies g(z) 1/a on Da (0). Letting a % 1, we obtain g(z) 1 on
D1 (a), which implies (6.10).
If f (z0 ) = z0  at some point in D1 (0), then g(z0 ) = 1, so Proposition 6.1 implies
g c, hence f (z) cz.
The next result is known as Liouvilles theorem.
Proposition 6.3. If f : C C is holomorphic and bounded, then f is constant.
Proof. Given z C, we have by (5.10), for each R (0, ),
Z
1
f ()
0
(6.11)
f (z) =
d,
2i
( z)2
DR (z)
where
(6.12)
DR (z) = { C :  z = R}.
66
(6.13)
1
=
2R
f (z + Reit )
iReit dt
R2 e2it
(6.14)
M
.
R
Compare the case n = 1 of (5.32). Since (6.13) holds for all R < , we obtain
f 0 (z) = 0,
(6.15)
z C,
f (z) a
z
0
f (0)
for z 6= 0,
for z = 0.
Then g : C C is holomorphic (cf. 5, Exercise 6). The hypothesis f (z) M for all z
implies
M + a
g(z)
for z = R,
R
so the maximum principle implies g(z) (M + a)/R on DR (0), and letting R
gives g 0, hence f a.
We are now in a position to prove the following result, known as the fundamental
theorem of algebra.
Theorem 6.4. If p(z) = an z n + an1 z n1 + + a1 z + a0 is a polynomial of degree n 1
(an 6= 0), then p(z) must vanish somewhere in C.
Proof. Consider
(6.16)
f (z) =
1
.
p(z)
If p(z) does not vanish anywhere on C, then f (z) is holomorphic on all of C. On the other
hand, when z 6= 0,
(6.17)
f (z) =
1
1
,
n
1
z an + an1 z + + a0 z n
67
so
(6.18)
f (z) 0, as z .
Thus f is bounded on C, if p(z) has no roots. By Proposition 6.3, f (z) must be constant,
which is impossible, so p(z) must have a complex root.
Alternatively, having (6.18), we can apply the maximum principle. Applied to f (z)
on DR (0), it gives f (z) sup=R f () for z R, and (6.18) then forces f to be
identically 0, which is impossible.
See Appendix F for an elementary proof of the fundamental theorem of algebra, i.e.,
a proof that does not make use of results from integral calculus.
Exercises
1. Establish the following improvement of Liouvilles theorem.
Proposition. Assume f : C C is holomorphic, and that there exist w0 C and a > 0
such that
f (z) w0  a, z C.
Then f is constant.
Hint. Consider
1
.
f (z) w0
g(z) =
1
(r) =
2
Z
0
f (p + rei ) d
68
satisfies
1
(r) =
2
(6.21)
f 0 (p + rei )ei d
1
=
2ir
Z
0
d
f (p + rei ) d.
d
p(z) = z n + an1 z n1 + + a1 z + a0 .
p(z) = (z r1 ) (z rn ).
6. Show that
(6.24)
p0 (z)
1
1
=
+ +
.
p(z)
z r1
z rn
7. Suppose each root rk of p(z) belongs to the right halfplane H = {z : Re z > 0}. Show
that
(6.25)
Re z < 0 = Re
p0 (z)
p0 (z)
< 0 =
6= 0.
p(z)
p(z)
8. Show that the set of zeros of p0 (z) is contained in the convex hull of the set of zeros of
p(z).
Hint. Given a closed set S Rn , the convex hull of S is the intersection of all the
halfspaces containing S.
9. Suppose f : C C is holomorphic and satisfies an estimate
f (z) C(1 + z)n1 ,
for some n Z+ . Show that f (z) is a polynomial in z of degree n 1.
Hint. Apply (5.32) with = DR (z) and let R to show that f (n) (z) = 0 for all z.
69
10. Show that if f : C C is holomorphic and not constant, then its range f (C) is dense
in C.
Hint. If f (C) omits a neighborhood of p C, consider the holomorphic function g(z) =
1/(f (z) p).
11. Consider the functions
f (z) = ez z,
f 0 (z) = ez 1.
Show that all the zeros of f are contained in {z : Re z > 0} while all the zeros of f 0 lie on
the imaginary axis. (Contrast this with the result of Exercise 7.)
Remark. Result of 29 imply that ez z has infinitely many zeros.
70
2f
2f
+
= 0,
x2
y 2
f =
2f
2f
+
+
.
x21
x2n
2u 2u
+ 2 = 0,
x2
y
2v
2v
+
= 0,
x2
y 2
so the real and imaginary parts of a function holomorphic on a region are both harmonic
on . Our first task in this section will be to show that many (though not all) domains
C have the property that if u C 2 () is real valued and harmonic, then there exists
a real valued function v C 2 () such that f = u + iv is holomorphic on . One says
v is a harmonic conjugate to u. Such a property is equivalent to the form (1.34) of the
CauchyRiemann equations:
(7.4)
u
v
=
,
x
y
v
u
= .
x
y
71
If also z , then Rz .
(7.6)
Z
u
u
v(z) =
dx +
dy
y
x
z
Z y
Z x
u
u
=
(a + is) ds
(t + iy) dt.
b x
a y
Also set
(7.7)
Z
u
u
v(z) =
dx +
dy
y
x
z
Z x
Z y
u
u
=
(t + ib) dt +
(x + is) ds.
a y
b x
v
u
(z) = (z),
x
y
and
(7.9)
v
u
(z) =
(z).
y
x
dx +
dy
y
x
Rz
(7.10)
ZZ 2
u 2u
=
+ 2 dx dy
x2
y
Rz
= 0,
the second identity by Greens theorem, (5.1), and the third because u = 0 on . Hence
(7.8)(7.9) give the CauchyRiemann equations (7.4), proving Proposition 7.1.
The next result, whose proof is similar to that of Proposition 1.10, simultaneously
extends the scope of Proposition 7.1 and avoids the use of Greens theorem.
72
If also z , then z .
(a + iy) +
(t + iy)
=
x
x
t=a
u
=
(z),
x
the second identity because u is harmonic and the third by the fundamental theorem of
calculus. This again establishes the CauchyRiemann equations, and completes the proof
of Proposition 7.2.
Later in this section, we will establish the existence of harmonic conjugates for a larger
class of domains. However, the results given above are good enough to yield some important
information on harmonic functions, which we now look into. The following is the mean
value property for harmonic functions.
Proposition 7.3. If u C 2 () is harmonic, z0 , and Dr (z0 ) , then
Z 2
1
(7.13)
u(z0 ) =
u(z0 + rei ) d.
2 0
Proof. We can assume u is real valued. Take > r such that D (z0 ) . By Proposition
7.1, u has a harmonic conjugate v on D (z0 ), so f = u + iv is holomorphic on D (z0 ). By
(5.8),
Z 2
1
(7.14)
f (z0 ) =
f (z0 + rei ) d.
2 0
Taking the real part gives (7.13).
As in (6.3), we also have, under the hypotheses of Proposition 7.3,
ZZ
Z 2 Z r
u(z0 + sei )s ds d
u(z) dx dy =
0
0
(7.15)
Dr (z0 )
= r2 u(z0 ),
73
hence
(7.16)
1
u(z0 ) =
Ar
ZZ
u(z) dx dy.
Dr (z0 )
Dr (q)
1
u(p) u(q) 2
r
ZZ
u(z) dx dy,
(p,q,r)
where
(7.21)
u(p) u(q)
4M p q
,
r
r < ,
74
1
f (z) + M + 1
is holomorphic on C and g(z) 1 for all z, so Proposition 6.3 implies g is constant, which
implies f is constant. (Compare Exercise 1 in 6.)
We return to the question of when does a harmonic function on a domain C have a
harmonic conjugate. We start with a definition. Let C be a connected, open set. We
say is a simply connected domain if the following property holds. Given p, q and a
pair of smooth paths
(7.24)
0 , 1 : [0, 1] ,
j (0) = p, j (1) = q,
s : [0, 1] ,
Compare material in Exercises 1013 of 5. The definition given there looks a little different, but it is equivalent to the one given here. We also write s (t) = (s, t), :
[0, 1] [0, 1] . We will prove the following.
Proposition 7.6. Let C be a simply connected domain. Then each harmonic function
u C 2 () has a harmonic conjugate.
The proof starts like that of Proposition 7.1, picking and setting
Z
u
u
(7.26)
v(z) =
dx +
dy ,
y
x
z
except this time, z denotes an arbitrary piecewise smooth path from to z. The crux
of the proof is to show that (7.26) is independent of the choice of path from to z. If this
known, we can simply write
Z z
u
u
(7.27)
v(z) =
dx +
dy ,
y
x
and proceed as follows. Given z , take r > 0 such that Dr (z) . With z = x + iy,
pick + iy, x + i Dr (z) (x, y, , R). We have
Z +iy
Z x+iy
u
u
u
u
(7.28)
v(z) =
dx +
dy +
dx +
dy ,
y
x
y
x
+iy
75
(7.29)
We also have
Z
(7.30)
x+i
v(z) =
u
u
dx +
dy +
y
x
x+iy
x+i
u
u
dx +
dy ,
y
x
v
u
(z) =
(z),
y
x
(7.32)
Then
Z
(7.33)
F1 dx + F2 dy
pq
F1 =
u
,
y
F2 =
u
,
x
F1
2u
= 2,
y
y
F2
2u
=
,
x
x2
76
is independent of s.
To prove Lemma 7.8, it is convenient to introduce some notation. Set
(7.37)
x1 = x,
x2 = y,
so (7.32) yields
Fk
Fj
=
,
xj
xk
(7.38)
=
Fj j0 .
F2
20
j
Then (7.36) takes the form
Z
(7.40)
0
We are assuming
: [0, 1] [0, 1] ,
(s, 0) = p,
(s, 1) = q,
F ((s, t))
(s, t) dt
t
0
Z 1X
(7.42)
j
Fj ((s, t))
=
(s, t) dt.
t
0
j
The sderivative of the integrand in (7.42) is obtained via the product rule and the chain
rule. Thus the sderivative of (7.42) is
Z 1X
Fj
77
(7.44)
to the second integrand in (7.43) and then integrate by parts. This involves applying /t
to Fj ((s, t)), and hence another application of the chain rule. When this is done, the
second integral in (7.43) becomes
Z 1X
Fj
(7.45)
Exercises
1. Modify the second proof of Proposition 7.5 to establish the following.
Proposition 7.5A. If u C 2 (C) is harmonic on C, real valued, and bounded from below,
then u is constant.
Hint. Let u = Re f, f : C C, holomorphic. See Exercise 1 of 6.
Alternative. Modify the argument involving (7.19)(7.23), used in the first proof of Proposition 7.5, to show that u(p) u(q) 0, and then reverse the roles of p and q.
2. Let u : R be harmonic. Assume u has a harmonic conjugate v, so f (z) = u(z)+iv(z)
is holomorphic. Show that, if : [a, b] is a piecewise smooth path,
Z
Z
Z
1
u
u
v
v
0
(7.46)
f (z) dz =
dx +
dy + i
dx +
dy .
i
y
x
y
x
Deduce that the integral on the right side of (7.46) must vanish whenever is a closed
curve in .
Hint. Write f 0 (z) dz = (ux + ivx )(dx + idy), separate the left side of (7.46) into real and
imaginary parts, and use the CauchyRiemann equations to obtain the right side of (7.46).
3. Let C be a connected, open set and u : R be harmonic. Show that u has a
harmonic conjugate v C 2 () if and only if
Z
u
u
(7.47)
dx +
dy = 0,
y
x
78
Z
u
u
dx +
dy = 4r2 0 (r2 ).
y
x
g() d,
z ,
pz
with pz a smooth path in from p to z. Show that this integral is independent of the
choice of such a path, so f (z) is well defined. Show that f is holomorphic on and
f 0 (z) = g(z),
z .
Hint. For the independence, see Exercise 14 of 5. For the rest, adapt the argument used
to prove Proposition 7.6, to show that
f
(z) = g(z),
x
and
1 f
(z) = g(z).
i y
79
Compare this conclusion with that of Lemma 7.7. Compare the relation between these
two results with the relation of Proposition 5.7 to Theorem 5.2.
10. Let C be open, and pq : [a, b] a path from p to q. Show that if v C 1 (),
then
Z
v
v
(7.49)
v(q) v(p) =
dx +
dy .
x
y
pq
80
f (z) =
Z
g() d =
g(s + ib) ds + i
a
(8.4)
f
(z) = ig(z).
y
z = z .
81
(8.6)
f (z),
z + L,
z .
Then g is holomorphic on .
Proof. It is readily verified that g is C 1 on and satisfies the CauchyRiemann equation
there, so g is holomorphic on \ L. Also it is clear that g is continuous on . To see that
g is holomorphic on all of , we show that g satisfies (8.1) whenever = R and R
is a (closed) rectangle. If R + or R this is clear. If R + L, it follows by
the continuity of g and a limiting argument (see Theorem 5.9); similarly we treat the case
R L. Finally, if R intersects both + and , then we set R = R+ R with
R = L, and note that
Z
Z
Z
g(z) dz,
g(z) dz =
g(z) dz +
R
R+
Exercises
1. Let C be a connected domain. Suppose is a smooth curve in , and \ has
two connected pieces, say . Assume g is continuous on , and holomorphic on + and
on . Show that g is holomorphic on .
Hint. Verify the hypotheses of Moreras theorem.
2. Suppose f is holomorphic on the semidisk z < 1, Im z > 0, continuous on its closure,
and real valued on the semicircle z = 1, Im z > 0. Show that setting
g(z) = f (z),
z 1, Im z > 0,
f (1/z),
1 z 2,
,
82
83
9. Goursats theorem
If C is open and f : C, we have defined f to be holomorphic provided
f C 1 () and f is complexdifferentiable. Here we show that the hypothesis f C 1 ()
can be dispensed with. This observation is due to E. Goursat.
Theorem 9.1. If f : C is complexdifferentiable at each point of , then f is
holomorphic, so f C 1 (), and in fact f C ().
Proof. We will show that the hypothesis yields
Z
(9.1)
f (z) dz = 0
R
Z
a
.
f
(z)
dz
4
R1
Divide R1 into four equal rectangles. One of them (call it R2 ) must have the property that
(9.3)
f (z) dz 42 a.
R2
(9.4)
f (z) dz 4k a.
Rk
with
(9.6)
84
In particular,
(9.7)
sup
zRk
(z)
= k 0, as k .
z p
Z
dz = 0,
Rk
z dz = 0.
Rk
Z
Z
f (z) dz =
(z) dz Ck 2k 2k ,
Rk
Rk
Exercise
1. Define f : R C by
1
f (x) = x2 sin , x 6= 0,
x
0,
x = 0.
1
f (z) = z 2 sin ,
z
z 6= 0.
85
f (z) =
an (z p)n .
n=0
If all an = 0, then f = 0 on DR (p). Otherwise, say aj is the first nonzero coefficient, and
write
(10.2)
f (z) = (z p) g(z),
g(z) =
aj+n (z p)n .
n=0
G(z) =
et
+tz
dt.
It is easy to see that the integral is absolutely convergent for each z C and defines a
continuous function of z. Furthermore, if is any closed curve on C (such as = R for
some rectangle R C) then we can interchange order of integrals to get
Z
(10.4)
Z
et
G(z) dz =
+tz
dz dt = 0,
the last identity by Cauchys integral theorem. Then Moreras theorem implies that G is
holomorphic on C.
86
For z = x real we can calculate (10.3) via elementary calculus. Completing the square
in the exponent gives
Z
Z
2
x2 /4
(tx/2)2
x2 /4
(10.5)
G(x) = e
e
dt = e
et dt.
(10.6)
R
0
I =
where
s2 t2
ds dt =
er r dr d = ,
(10.7)
G(x) =
/4
x R.
/4
z C,
ex
, so
G(z) =
ez
2
2
(10.9)
et +ity dt = ey /4 , y R.
ey
/4
eg(z) = f (z),
z .
Proof. We may as well assume f is not constant. Take p such that f (p)
/ R =
(, 0]. Let O be a neighborhood of p such that
(10.11)
f : O C \ R .
We can define g on O by
(10.12)
z O,
where log : C \ R C is as in 4. Applying d/dz and using the chain rule gives
(10.13)
g 0 (z) =
f 0 (z)
,
f (z)
z O.
87
Consequently
Z
(10.14)
f 0 ()
d,
f ()
eg(z) = f (z),
z O,
z ,
These functions are hence analytically continued from O (where they have a standard
definition from 4) to . We will see further cases of analytic continuation in Sections 18
and 19.
Exercises
1. Suppose C is a connected region that is symmetric about the real axis, i.e.,
z z . If f is holomorphic on and real valued on R, show that
(10.19)
f (z) = f (z).
88
is reflection about the real axis. Suppose C is a connected region that is symmetric
about the imaginary axis, i.e., z z . If f is holomorphic on and is purely
imaginary on iR, show that
f (z) = f (z ) .
(10.20)
d g(z)
e
f (z) = 0,
dz
and deduce that (10.10) holds, without appealing to Proposition 10.1.
5. Consider
Z
I(a) =
eat dt,
Re a > 0.
1/2
I(a) =
a
.
2
Hint. Use a change of variable to evaluate I(a) for a (0, ).
6. Evaluate
b > 0.
i/4
lim I(b i) =
e
.
b&0
2
R
R
What does this say about 0 cos t2 dt and 0 sin t2 dt?
89
11. Singularities
The function 1/z is holomorphic on C \ 0 but it has a singularity at z = 0. Here we will
make a further study of singularities of holomorphic functions.
A point p C is said to be an isolated singularity of f if there is a neighborhood U of p
such that f is holomorphic on U \ p. The singularity is said to be removable if there exists
a holomorphic function f on U such that f = f on U \ p. Clearly 0 is not a removable
singularity for 1/z, since this function is not bounded on any set D (0) \ 0. This turns out
to be the only obstruction to removability, as shown in the following result, known as the
removable singularities theorem.
Theorem 11.1. If p and f is holomorphic on \p and bounded, then p is a removable
singularity.
Proof. Consider the function g : C defined by
g(z) = (z p)2 f (z),
(11.1)
z \ p,
g(p) = 0.
z \ p,
g (p) = 0.
g(z) =
an (z p)n ,
z U.
n=0
h(z) =
a2+n (z p)n ,
z U.
n=0
f(z) = f (z),
z \ p,
f(p) = h(p)
90
X
(11.6)
g(z) =
an (z p)n ,
n=k
where we have picked ak as the first nonzero coefficient in the power series. Hence
(11.7)
h(p) = ak 6= 0,
f (z) = e1/z ,
for which 0 is an essential singularity. The following result is known as the CasoratiWeierstrass theorem.
Proposition 11.3. Suppose f : \ p C has an essential singularity at p. Then, for
any neighborhood U of p in , the image of U \ p under f is dense in C.
Proof. Suppose that, for some neighborhood U of p, the image of U \ p under f omits a
neighborhood of w0 C. Replacing f (z) by f (z) w0 , we may as well suppose w0 = 0.
Then g(z) = 1/f (z) is holomorphic and bounded on U \ p, so p is a removable singularity
for g, which hence has a holomorphic extension g. If g(p) 6= 0, then p is removable for f .
If g(p) = 0, then p is a pole of f .
Remark. There is a strengthening of Proposition 11.3, due to E. Picard, which we will
treat in 28.
A function holomorphic on except for a set of poles is said to be meromorphic on .
For example,
sin z
tan z =
cos z
is meromorphic on C, with poles at {(k + 1/2) : k Z}.
Here we have discussed isolated singularities. In 4 we have seen examples of functions,
such as z 1/2 and log z, with singularities of a different nature, called branch singularities.
Another useful consequence of the removable singularities theorem is the following characterization of polynomials.
91
g(z) = f
1
z
(11.11)
G(z) =
k1
X
gj z j + z k h(z),
j=0
k1
X
gj z jk + h(z),
j=0
so
f (z) =
k1
X
gj z kj + h
1
z
j=0
for z 6= 0.
It follows that
(11.13)
f (z)
k1
X
gj z kj
j=0
is holomorphic on C and, as z , this tends to the finite limit h(0). Hence, by
Liouvilles theorem, this difference is constant, so f (z) is a polynomial.
Exercises
1. In the setting of Theorem 11.1, suppose z p = r and Dr (z) . If f  M on \ p,
show that
(11.14)
f 0 (z)
M
,
r
92
Use this to show directly that the hypotheses of Theorem 11.1 imply g C 1 (), avoiding
the need for Goursats theorem in the proof.
Hint. Use (5.10) to prove (11.14). In (5.10), replace by Ds (z) with s < r, and then let
s % r.
2. For yet another approach to the proof of Theorem 11.1, define h : C by
(11.16)
z \ p,
h(p) = 0.
R
Show that h : C is continuous. Show that R h(z) dz = 0 for each closed rectangle
R , and deduce that h is holomorphic on . Use a power series argument parallel to
that of (11.3)(11.4) to finish the proof of Theorem 11.1.
3. Suppose , O C are open and f : O is holomorphic and a homeomorphism.
Show that f 0 (p) 6= 0 for all p .
Hint. Apply the removable singularities theorem to f 1 : O .
Compare the different approach suggested in Exercise 8 of 5.
4. Let f : C C be holomorphic and assume f is not a polynomial. (We say f is a
transcendental function.) Show that g : C \ 0 C, defined by g(z) = f (1/z), has an
essential singularity at 0. Apply the CasoratiWeierstrass theorem to g, and interpret the
conclusion in terms of the behavior of f .
Remark. (Parallel to that after Proposition 11.3) For an improvement of this conclusion,
see 28.
93
A = {z C : r0 < z z0  < r1 }.
be such an annulus. For now assume 0 < r0 < r1 < . Let j be the counterclockwise
circles {z z0  = rj }, so A = 1 0 . If f C 1 (A) is holomorphic in A, the Cauchy
integral formula gives
1
f (z) =
2i
(12.2)
Z
1
f ()
1
d
z
2i
f ()
d,
z
1 ,
1
1
,
z0
z z0 1 zz
0
0 ,
z z0  <  z0 ,
for 1 ,
>  z0 ,
for 0 ,
to write
(12.3)
1
1 X z z0 n
=
,
z
z0 n=0 z0
1 ,
and
(12.4)
1
1 X z0 m
=
,
z
z z0 m=0 z z0
0 .
f (z) =
X
n=
an (z z0 )n ,
z A,
94
with
(12.6)
1
an =
2i
Z
1
f ()
d,
( z0 )n+1
n 0,
and
(12.7)
1
an =
2i
Z
f ()( z0 )m d,
n = m 1 < 0.
Now in (12.6) and (12.7) 0 and 1 can be replaced by any circle in A concentric with
these. Using this observation, we can prove the following.
Proposition 12.1. Given 0 r0 < r1 , let A be the annulus (12.1). If f : A C is
holomorphic, then it is given by the absolutely convergent series (12.5), with
(12.8)
1
an =
2i
f ()
d,
( z0 )n+1
n Z,
f (z) =
1
,
z1
on the regions
(12.10)
{z : z < 1}
95
X
1
f (z) =
=
zk ,
1z
(12.11)
k=0
1 1
f (z) =
z 1
1
z
1
X
1 X 1 k
zk .
=
=
z
z
k=0
k=
Similarly, for
(12.13)
g(z) =
1
z2
on the regions
(12.14)
{z : z < 2}
1 X z k
=
,
2
2
1 1
g(z) =
2 1
z
2
k=0
g(z) =
1 1
z 1
2
z
1
X
1 X 2 k
=
2k1 z k .
z
z
k=0
k=
Exercises
Consider the Laurent series
ez+1/z =
(12.17)
an z n .
n=
1
an =
Z
0
96
1
.
m!(m
+
k)!
m=0
Hint. For Exercise 2, use (12.7); for Exercise 3 use the series for ez and e1/z .
4. Consider the function
f (z) =
1
.
(z 1)(z 2)
an (z z0 )n ,
n=
valid on {z : 0 < z z0  < b} for some b > 0. Use (12.8), letting shrink, to show that
an = 0 for each n < 0, thus obtaining another proof of the removable singularities theorem
(Theorem 11.1).
6. Show that, for z sufficiently small,
(12.18)
1 ez + 1
1 X
=
+
ak z 2k1 .
2 ez 1
z
k=1
1 cosh z/2
,
2 sinh z/2
F 0 (z) =
1
F (z)2 .
4
Using (2.19)(2.20), write out the Laurent expansion for F (z)2 , in terms of that for F (z)
given above. Comparing terms in the expansions of F 0 (z) and 1/4 F (z)2 , show that
a1 =
1
,
2
97
and, for k 2,
k1
1 X
ak =
a` ak` .
2k + 1
`=1
Bk
,
(2k)!
X
X
z
z n X z `
2+
=
1+
ak z 2k ,
2
n!
`!
n=1
`=1
k=1
X
z `
`=1
`!
z X
1 +
ak z 2k .
2
k=1
98
f () =
f(k) eik ,
k=
1 if ` = 0,
we see that
(13.3)
1
f(k) =
2
f ()eik d.
The series on the right side of (13.1) is called the Fourier series of f .
If f is given by (13.3), to examine whether (13.1) holds, we first sneak up on the sum
on the right side. For 0 < r < 1, set
(13.4)
Jr f () =
f(k)rk eik .
k=
Note that
(13.5)
1
f(k)
2
f () d,
so whenever the right side of (13.5) is finite we see that the series (13.4) is absolutely
convergent for each r [0, 1). Furthermore, we can substitute (13.3) for f in (13.4) and
change order of summation and integration, to obtain
Z
(13.6)
Jr f () = f (0 )pr ( 0 ) d0 ,
S1
where
1 X k ik
pr () = p(r, ) =
r e
2
(13.7)
=
k=
h
X
1
1+
2
i
(rk eik + rk eik ) ,
k=1
99
p(r, ) =
S1
we see that, for r close to 1, p(r, ) as a function of is highly peaked near = 0 and small
elsewhere, as in Fig. 13.1. Given these facts, the following is an exercise in real analysis.
Proposition 13.1. If f C(S 1 ), then
(13.10)
Jr f f uniformly on S 1 as r % 1.
Jr f () =
f ( 0 )pr (0 ) d0 .
[f ( 0 ) f ()]pr (0 ) d0
Z
+
f ( 0 )pr (0 ) d0
+
(13.13)
 0 
= I + II + III.
We have
I = f ()(1 (r, )),
II sup f ( 0 ) f (),
(13.14)
 0 
100
Proposition 13.2. Assume f C(S 1 ). If the Fourier coefficients f(k) form a summable
series, i.e., if
(13.15)
f(k) < ,
k=
ak = S = lim
r%1
rk ak = S.
ak  < .
k>N
Then
N
X
SN =
ak = S SN  < ,
k=N
and
k
r ak < ,
r (0, 1).
k>N
Since clearly
lim
r%1
N
X
k=N
k
r ak =
N
X
ak ,
k=N
101
hence
Z
1
f(k)
2k
(13.17)
f 0 () d.
1
f(k) =
2k 2
hence
f 00 ()eik d,
1
f(k)
2k 2
f 00 () d.
f(k)
2(k 2 + 1)
S1
Hence
(13.20)
f C 2 (S 1 ) =
f(k) < .
f () = ,
f (k) =
 eik d
2
(13.22)
1
= 1 (1)k
,
k 2
for k 6= 0, while f(0) = /2. It is clear this forms a summable series, so Proposition 13.2
implies that, for ,
 =
(13.23)
=
X 2
eik
2
k 2
k odd
`=0
1
cos(2` + 1).
(2` + 1)2
102
(13.24)
`=0
1
2
=
.
(2` + 1)2
8
Writing
X
1
k2
(13.25)
k=1
as a sum of two series, one for k 1 odd and one for k 2 even, yields an evaluation of
(13.25). (See Exercise 1 below.)
We see from (13.23) that if f is given by (13.21), then f(k) satisfies
f(k)
(13.26)
k2
C
.
+1
f (k) =
f 0 ()
(e
) d
2ik
k
(13.27)
Z
i
1 h 00
ik
0
0
=
f
()e
d
+
f
()
f
()
,
2k 2
which yields (13.26).
Given f C(S 1 ), let us say
(13.28)
f A(S 1 )
f(k) < .
Proposition 13.2 applies to elements of A(S 1 ), and Proposition 13.3 gives a sufficient
condition for a function to belong to A(S 1 ). A more general sufficient condition will be
given in Proposition 13.6.
R
We next make use of (13.2) to produce results on S 1 f ()2 d, starting with the following.
103
1
f(k)2 =
2
Z
f ()2 d.
S1
1
f(k)
g (k) =
2
Z
f ()g() d.
S1
Proof. Switching order of summation and integration and using (13.2), we have
Z
Z X
1
1
f ()g() d =
f(j)
g (k)ei(jk) d
2
2
S1
S 1 j,k
(13.31)
X
=
f(k)
g (k),
k
Given f and f 2 integrable on S 1 (we say f is square integrable), this defines a norm,
satisfying the following result, called the triangle inequality:
(13.34)
See Appendix H for details on this. Behind these results is the fact that
(13.35)
104
(13.37)
f(k)2 = kf k2L2 ,
f(k)
g (k) = (f, g)L2 .
The left side of (13.37) is the square norm of the sequence (f(k)) in `2 . Generally, a
sequence (ak ) (k Z) belongs to `2 if and only if
(13.39)
k(ak )k2`2 =
ak 2 < .
((ak ), (bk )) =
ak bk .
f f in L2 kf f kL2 0.
f A(S 1 ) = SN f f in L2 , as N .
(13.45)
kf kL2 kSN f kL2 kf SN f kL2 ,
and clearly (by Proposition 13.4)
(13.46)
kSN f k2L2
N
X
k=N
f(k)2 ,
105
so
kf SN f kL2 0 as N = kf k2L2 =
(13.47)
f(k)2 .
We now consider more general square integrable functions f on S 1 . With f(k) and SN f
defined by (13.3) and (13.32), we define RN f by
(13.48)
Note that
f = SN f + RN f.
R
S1
f ()eik d =
(13.49)
S1
and hence
(13.50)
(SN f, RN f )L2 = 0.
Consequently,
(13.51)
In particular,
(13.52)
lim kf f kL2 = 0,
lim kf SN f kL2 = 0.
Then
(13.55)
lim kf SN f kL2 = 0.
106
(13.58)
Lemma 13.5 also applies to many discontinuous functions. Consider, for example
f () = 0 for < < 0,
1 for 0 < < .
(13.59)
We can set, for N,
f () = 0
for 0,
1
for 0 ,
1
1
1 for
1
( ) for .
(13.60)
Then each f C(S 1 ). (In fact, f A(S 1 ), by Proposition 13.3.). Also, one can check
that kf f k2L2 2/. Thus the conclusion in (13.58) holds for f given by (13.59).
More generally, any piecewise continuous function on S 1 is an L2 limit of continuous
functions, so the conclusion of (13.58) holds for them. To go further, let us recall the class
of Riemann integrable functions. (Details can be found in Chapter 4, 2 of [T0] or in 0 of
[T].) A function f : S 1 R is Riemann integrable provided f is bounded (say f  M )
and, for each > 0, there exist piecewise constant functions g and h on S 1 such that
Z
(13.61)
g f h ,
and
h () g () d < .
S1
Then
Z
(13.62)
Z
f () d = lim
S1
0
S1
Z
g () d = lim
0
S1
h () d.
107
S1
S1
(13.63)
M +1
<
,
This is not the end of the story. There are unbounded functions on S 1 that are square
integrable, such as
1
(13.65)
f () =  on [, ], 0 < < .
2
In such a case, one can take f () = min(f (), ), N. Then each f is continuous and
kf f kL2 0 as . Hence the conclusion of (13.64) holds for such f .
The ultimate theory of functions for which the result
(13.66)
SN f f in L2 norm
holds was produced by H. Lebesgue in what is now known as the theory of Lebesgue
measure and integration. There is the notion of measurability
of a function f : S 1
R
2
1
C. One says f L (S ) provided f is measurable and S 1 f ()2 d < , the integral
here being the Lebesgue integral. Actually,
L2 (S 1 ) consists of equivalence classes of such
R
functions, where f1 f2 if and only if f1 () f2 ()2 d = 0. With `2 as in (13.39), it
is then the case that
(13.67)
F : L2 (S 1 ) `2 ,
given by
(13.68)
f L2 (S 1 ),
and
(13.70)
SN f f in L2 ,
f L2 (S 1 ).
For the reader who has not seen Lebesgue integration, we refer to books on the subject
(eg., [T3]) for more information.
We mention two key propositions which, together with the arguments given above,
establish these results. The fact that Ff `2 for all f L2 (S 1 ) and (13.69)(13.70) hold
follows via Lemma 13.5 from the following.
108
ak eik
k
kf f k2L2 =
ak 2 0 as .
<k
That is to say, (f ) is a Cauchy sequence in L2 (S 1 ). Surjectivity follows from the fact that
Cauchy sequences in L2 (S 1 ) always converge to a limit:
Proposition B. If (f ) is a Cauchy sequence in L2 (S 1 ), there exists f L2 (S 1 ) such
that f f in L2 norm.
Proofs of these results can be found in the standard texts on measure theory and integration, such as [T3].
We now establish a sufficient condition for a function f to belong to A(S 1 ), more general
than that in Proposition 13.3.
P
Proposition 13.6. If f is a continuous, piecewise C 1 function on S 1 , then
f (k) < .
Proof. As in the proof of Proposition 13.3, we can reduce the problem to the case f
C 1 ([, ]), f () = f (). In such a case, with g = f 0 C([, ]), the integration by
parts argument (13.16) gives
1
f(k) = g(k),
ik
(13.71)
k 6= 0.
By (13.64),
X
(13.72)

g (k)2 = kgk2L2 .
k6=0
f(k)
X 1 1/2 X
1/2
2

g (k)
k2
k6=0
k6=0
CkgkL2 .
This completes the proof.
There is a great deal more that can be said about convergence of Fourier series. For example, material presented in the appendix to the next section has an analogue for Fourier
109
series. We also mention Chapter 5, 4 in [T0]. For further results, one can consult treatments of Fourier analysis such as Chapter 3 of [T2].
Fourier series connects with the theory of harmonic functions, as follows. Taking z =
rei in the unit disk, we can write (13.4) as
(13.74)
Jr f () =
f(k)z k +
k=0
f(k)z k .
k=1
We write this as
(13.75)
1 z2
PI f (z) =
2
f (w)
ds(w),
w z2
S1
0
the integral being with respect to arclength on S 1 . To see this, note that if w = ei and
z = rei , then ds(w) = d0 and
0
= 1 r(ei( ) + ei( ) ) + r2 .
Since solutions to u = 0 remain solutions upon translation and dilation of coordinates,
we have the following result.
Proposition 13.7. If D C is an open disk and f C(D) is given, there exists a
unique u C(D) C 2 (D) satisfying
(13.77)
u = 0 on D,
uD = f.
110
v(z) = u(z),
u(z),
z + L,
z .
Then v is harmonic on .
Proof. It is readily verified that v is harmonic in + and continuous on . We
need to show that v is harmonic on a neighborhood of each point p L. Let D = Dr (p)
be a disk centered at p such that D . Let f C(D) be given by f = vD . Let
w C 2 (D) C(D) be the unique harmonic function on D equal to f on D.
Since f is odd with respect to reflection about the real axis, so is w, so w = 0 on D R.
+
Thus both v and w are harmonic on D+ = D {Im z > 0}, and continuous on D , and
+
g(z) = f (z),
z + ,
f (z),
z .
is holomorphic. Now Im h = Im f on + , so g h is real valued on + , so, being holomorphic, it must be constant. Thus, altering w by a real constant, we have
(13.81)
h(z) = g(z),
z + .
h(z) = h(z),
z .
111
z + ,
h(z) = g(z),
Exercises
1. Verify the evaluation of the integral in (13.22). Use the evaluation of (13.23) at = 0
(as done in (13.24)) to show that
(13.84)
X
2
1
=
.
k2
6
k=1
X
4
1
=
.
k4
90
k=1
X
f (z) =
an z n .
n=
Show that
an = g(n),
g = f S 1 .
112
6. Show that
h = f g = h(k)
= f(k)
g (k).
7. Show that
2
f, g L (S ), h = f g =
h(k)
< .
k=
(13.80)
ei + z
Re i
f () d.
e z
P
k
k Z.
13. Let f C(S 1 ) be real valued. Show that PI f and (1/i) PI g are harmonic conjugates,
provided
g(k) = f(k)
for k > 0,
f(k) for k < 0.
113
(14.2)
f () = Ff () =
f (x)eix dx,
2
R.
Similarly, we set
1
F f () =
2
(14.2A)
f (x)eix dx,
R,
(14.3)
f ()
f (x) dx.
2
We also have continuity.
Proposition 14.1. If f is integrable on R, then f is continuous on R.
R
Proof. Given > 0, pick N < such that x>N f (x) dx < . Write f = fN + gN ,
where fN (x) = f (x) for x N, 0 for x > N . Then
f() = fN () + gN (),
(14.4)
and

gN () < ,
2
(14.5)
Meanwhile, for , R,
(14.6)
1
fN () fN () =
2
and
(14.6A)
x  
N  ,
114
for x N , so
N
fN () fN () kf kL1  ,
2
(14.7)
where
1
H (x) =
2
(14.11)
+ix
d.
A change of variable shows that H (x) = (1/ )H1 (x/ ), and the computation of
H1 (x) is accomplished in 10; we see that H1 (x) = (1/2)G(ix), with G(z) defined by
(10.3) and computed in (10.8). We obtain
H (x) =
(14.12)
The computation
ex dx =
2
1
ex /4 .
4
(14.13)
H (x) dx = 1,
> 0.
We see that H (x) is highly peaked near x = 0 as & 0. An argument similar to that
used to prove Proposition 13.1 then establishes the following.
115
J f (x) F f(x),
as 0.
where Riemann integrability on [N, N ] has a definition similar to that given in (13.61)
(13.62) for functions on S 1 . Still more general is Lebesgues class, consisting of measurable
functions f : R C satisfying (14.1), where the Lebesgue integral is used. An element of
1
we say f1 f2 provided
RL(R) consists of an equivalence class of such functions, where
1
f f2  dx = 0. The quantity kf kL1 is called the L norm of f . It satisfies the
1
triangle inequality
(14.16)
116
Proposition B1. If (f ) is a Cauchy sequence in L1 (R), there exists f L1 (R) such that
f f in L1 .
As in 13, we will also be interested in square integrable functions. A function f : R C
is said to be square integrable if f and f 2 are integrable on each finite interval [N, N ]
and
Z
2
(14.17)
kf kL2 =
f (x)2 dx < .
Taking the square root gives kf kL2 , called the L2 norm of f . Parallel to (13.34) and
(14.16), there is the triangle inequality
(14.18)
The proof of (14.18) is not as easy as that of (14.16), but, like (13.34), it follows, via results
of Appendix H, from the fact that
(14.19)
(f, g)L2 =
f (x)g(x) dx.
The triangle inequality (14.18) makes L2 (R) a metric space, with distance function d(f, g) =
kf gkL2 , and we say f f in L2 if kf f kL2 0. Parallel to Propositions A1 and
B1, we have the following.
Proposition A2. Given f L2 (R) and k N, there exist f C0k (R) such that f f
in L2 .
Proposition B2. If (f ) is a Cauchy sequence in L2 (R), there exists f L2 (R) such that
f f in L2 .
As in 13, we refer to books on measure theory, such as [T3], for further material on
L (R) and L2 (R), including proofs of results stated above.
Somewhat parallel to (13.28), we set
1
(14.21)
By Corollary 14.3, the Fourier inversion formula (14.8) holds for all f A(R). It also
follows that f A(R) f A(R). Note also that
(14.22)
A(R) L2 (R).
117
In fact, if f A(R),
Z
kf k2L2
(14.23)
f (x)2 dx
Z
sup f (x) f (x) dx
=
1
kfkL1 kf kL1 .
2
It is of interest to know when f A(R). We mention one simple result here. Namely,
if f C k (R) has compact support (we say f C0k (R)), then integration by parts yields
(14.24)
0 j k.
Hence
(14.25)
While (14.25) is crude, it will give us a start on the L2 theory of the Fourier transform.
Let us denote by F the map f 7 f, and by F the map you get upon replacing eix by
eix . Then, with respect to the inner product (14.20), we have, for f, g A(R),
(14.26)
for f, g A(R).
The result (14.28) is called the Plancherel identity. Using it, we can extend F and F
to act on L2 (R), obtaining (14.28) and the Fourier inversion formula on L2 (R).
Proposition 14.4. The maps F and F have unique continuous linear extensions from
(14.28A)
F, F : A(R) A(R)
to
(14.29)
F, F : L2 (R) L2 (R),
118
F Ff = f,
FF f = f
The argument goes like this. Given f L2 (R), take f A(R) such that f f in L2 .
Then kf f kL2 0 as , . Now (14.28), applied to f f A(R), gives
(14.32)
From here, the result (14.30) and the extension of (14.28) to L2 (R) follow.
Given f L2 (R), we have
[R,R] f f in L2 ,
as R ,
1
SR f (x) =
2
f()eix d.
Then
(14.35)
f L2 (R) = SR f f in L2 (R),
as R .
Having Proposition 14.4, we can sharpen (14.25) as follows.
119
(14.36)
(14.37)
Proposition 14.4 implies
kf + gkL2 = kf + f 0 kL2 .
(14.38)
(1 + i)1 (1 + i)f() d
kfkL1 =
nZ
Z
o1/2
d o1/2 n
()2 d
(1
+
i)
f
1 + 2
= kf + f 0 kL2 ,
(14.39)
the inequality in (14.39) by Cauchys inequality (cf. (H.18)) and the last identity by (14.37)
(14.38). This proves (14.36).
Remark. Parallel to Proposition 13.6, one can extend Proposition 14.6 to show that if
f has compact support, is continuuous, and is piecewise C 1 on R, then f A(R). In
conjunction with (14.39), the following is useful for identifying other elements of A(R).
Proposition 14.7. Let f A(R) and f C(R) L1 (R). Assume
f f in L1 norm, and kf kL1 A,
for some A < . Then f A(R).
Proof. Clearly f f uniformly on R. Hence, for each R < ,
Z
f () f() d 0,
as .
Thus
f() d A,
R < ,
120
The interested reader can consult 14A for a still sharper condition guaranteeing that
f A(R).
Exercises
In Exercises 12, assume f : R C is a C 2 function satisfying
f (j) (x) C(1 + x)2 ,
(14.40)
j 2.
1. Show that
f()
(14.41)
C0
,
2 + 1
R.
X
X
1
ikx
f (k)e
=
f (x + 2`).
2 k=
`=
(14.42)
3. Use f (x) = ex
(14.43)
/4t
X
1 X k2 /
`2
e
=
e
.
`=
k=
2
Hint. Use (14.11)(14.12) to get f() = 2t et . Take t = , and set x = 0 in (14.42).
4. For each of the following functions f (x), compute f().
(c)
f (x) = ex ,
1
f (x) =
,
1 + x2
f (x) = [1/2,1/2] (x),
(d)
(a)
(b)
121
5. In each case, (a)(d), of Exercise 4, record the identity that follows from the Plancherel
identity (14.28).
Exercises 68 deal with the convolution of functions on R, defined by
Z
(14.44)
f g(x) =
f (y)g(x y) dy.
6. Show that
kf gkL1 kf kL1 kgkL1 ,
7. Show that
\
(f
g)() =
2 f()
g ().
8. Compute f f when f (x) = [1/2,1/2] (x), the characteristic function of the interval
[1/2, 1/2]. Compare the result of Exercise 7 with the computation of (d) in Exercise 4.
9. Prove the following result, known as the RiemannLebesgue lemma.
(14.45)
Hint. (14.41) gives the desired conclusion for f when f C02 (R). Then use Proposition
A1 and apply (14.3) to f f .
10. Sharpen the result of Exercise 1 as follows, using the reasoning in the proof of Proposition 14.6. Assume f : R C is a C 1 function satisfying
f (j) (x) C(1 + x)2 ,
j 1.
fh (x) = f (x + h).
122
Proposition 14.8. If f L1 (R) L2 (R) and there exists C < such that
kf fh kL2 Ch ,
(14.47)
h [0, 1],
with
(14.48)
>
1
,
2
then f A(R).
Proof. A calculation gives
fh () = eih f(),
(14.49)
so, by the Plancherel identity,
Z
(14.50)
kf
fh k2L2
1 eih 2 f()2 d.
Now,
3
h
= 1 eih 2 2,
2
2
(14.51)
so
(14.52)
kf
fh k2L2
f()2 d.
3
2 h 2
f()2 d C22` .
2` 2`+1
n
(14.55)
f()2 d
2` 2`+1
C2` 2`/2
= C2(1/2)` .
o1/2
1 d
2` 2`+1
o1/2
123
(14.57)
if 0 x 1,
otherwise.
We have, for 0 h 1,
kf fh k2L2 = 2h,
(14.58)
so (14.47) holds, with = 1/2. Since A(R) C(R), this function does not belong to
A(R), so the condition (14.48) is about as sharp as it could be.
Remark. Using
(14.59)
Z
gh dx sup g
h dx,
so, with
(14.61)
kf kC r =
sup
xR,0<h1
hr f (x) fh (x),
f L2 (R), f = 0 = f = 0.
We call these Fourier uniqueness results. An extension of (14.63) is the following consequence of Proposition 14.2:
(14.65)
f L1 (R) C(R), f = 0 = f = 0.
Here, we advertize the following strengthening of (14.61).
124
(14.66)
We indicate a proof of this result, starting with the following variant of (14.26). If
f L1 (R) and also g L1 (R), then
Z
(14.67)
Z Z
1
f ()g() d =
f (x)eix g() dx d
2
Z
=
f (x)
g (x) dx,
where the second identity uses a change in the order of integration. Thus
Z
(14.68)
f L (R), f = 0 =
1
f (x)h(x) dx = 0,
for all h = g, g L1 (R). In particular, (14.64) holds for all h A(R), and so, by (14.25),
it holds for all h C02 (R). The implication
Z
(14.69)
f L (R),
125
f (t)est dt,
Lf (s) =
Re s > A.
It is clear that this integral is absolutely convergent for each s in the halfplane HA = {z
C : Re z > A} and defines a continuous function Lf : HA C. Also, if is a closed curve
(e.g., the boundary of a rectangle) in HA , we can change order of integration to see that
Z
(15.3)
Z
f (t)est ds dt = 0.
Lf (s) ds =
0
(15.5)
126
with
(15.10)
Lf (s) =
Lg(s) + q(s)
,
p(s)
(15.12)
Lf (B + i) = 2 (),
R,
with
(15.13)
(x) = f (x)eBx ,
0
x 0,
x < 0.
d,
2
hence giving, for t > 0,
(15.15)
Z
eBt
f (t) =
Lf (B + i)eit d
2
Z
1
=
Lf (s)est ds,
2i
127
Proposition 15.1. If f1 and f2 are integrable on [0, R] for all R < and satisfy (15.1),
then
(15.16)
s HA = f1 = f2 on R+ .
We can also use material of 10 to deduce that f1 = f2 given Lf1 (s) = Lf2 (s) on a set
with an accumulation point in HA .
Exercises
1. Show that the Laplace transform of f (t) = tz1 , for Re z > 0, is given by
Lf (s) = (z) sz ,
(15.17)
where (z) is the Gamma function:
Z
(15.18)
(z) =
et tz1 dt.
eat ,
(b)
cosh at,
(c)
sinh at,
(d)
sin at,
(e)
tz1 eat .
3. Compute the inverse Laplace transforms of the following functions (defined in appropriate right halfspaces).
(a)
(b)
(c)
(d)
(e)
1
,
sa
s
,
2
s a2
a
,
2
s a2
a
,
2
s + a2
1
.
s+1
128
f g(t) =
f ( )g(t ) d.
0
4. Show that (15.19) coincides with the definition (14.44) of convolution, provided f (t)
and g(t) vanish for t < 0.
5. Show that if fz (t) = tz1 for t > 0, with Re z > 0, and if also Re > 0, then
(15.20)
with
Z
B(z, ) =
6. Show that
(15.21)
129
f (z) =
an (pj )(z pj )n .
n=
f (z) dz =
X Z
f (z) dz = 2i
j D
j
Respj (f ).
This identity provides a useful tool for computing a number of interesting integrals of
functions whose antiderivatives we cannot write down. Examples almost always combine
the identity (16.3) with further limiting arguments. We illustrate this method of residue
calculus to compute integrals with a variety of examples.
To start with a simple case, let us compute
Z
(16.4)
dx
.
1 + x2
In this case we can actually write down an antiderivative, but never mind. The function
f (z) = (1 + z 2 )1 has simple poles at z = i. Whenever a meromorphic function f (z) has
a simple pole at z = p, we have
(16.5)
In particular,
(16.6)
Resi (1 + z 2 )1 = lim (z i)
zi
1
1
= .
(z + i)(z i)
2i
130
Let R be the path formed by the path R from R to R on the real line, followed by the
semicircle R of radius R in the upper halfplane, running from z = R to z = R. Then
R encloses the pole of (1 + z 2 )1 at z = i, and we have
Z
dz
= 2i Resi (1 + z 2 )1 = ,
1 + z2
(16.7)
R
provided R > 1. On the other hand, considering the length of R and the size of (1 + z 2 )1
on R , it is clear that
Z
(16.8)
lim
R
R
dz
= 0.
1 + z2
Hence
Z
(16.9)
dx
= lim
R
1 + x2
dz
= .
1 + z2
This is consistent with the result one gets upon recalling that d tan1 x/dx = (1 + x2 )1 .
For a more elaborate example, consider
Z
(16.10)
dx
.
1 + x4
p1 = ei/4 ,
p2 = e3i/4 ,
p3 = e3i/4 ,
p4 = ei/4 .
Resp1 (1 + z 4 )1 =
(16.12)
Resp2 (1 + z 4 )1
dx
= lim
R
1 + x4
2
X
dz
= 2i
Respj (1 + z 4 )1
1 + z4
j=1
i 3i/4
(e
+ ei/4 ) = ,
2
2
131
(16.14)
(16.15)
eix
dx,
1 + x2
for R. Note that f (z) = eiz /(1 + z 2 ) has simple poles at z = i, and
(16.16)
eiz
e
=
.
1 + z2
2i
Resi
(16.17)
eix
dx = e ,
1 + x2
0.
For < 0 one can make a similar argument, replacing R by its image R reflected across
the real axis. Alternatively, we can see directly that (16.15) defines an even function of ,
so
Z
eix
(16.18)
dx = e , R.
2
1
+
x
The reader can verify that this result is consistent with the computation of
Z
ex eix dx
A=
eix
2 cosh
x
2
dx.
To evaluate this, we compare it with the integral over the path (x) = x 2i. See
Fig. 16.1. We have
Z
(16.20)
eiz
dz =
2 cosh z2
e2+ix
dx = e2 A,
2 cosh x2
132
since cosh(y i) = cosh y. Now the integrand has a pole at z = i, and a computation
gives
(16.21)
Resi
eiz
= i e .
2 cosh z2
We see that
A e2 A = 2i Resi
(16.22)
eiz
,
2 cosh z2
and hence
Z
(16.23)
eix
.
x dx =
2 cosh 2
cosh
The evaluation of (16.19) involved an extra wrinkle compared to the other examples
given above, involving how the integrand on one path is related to the integrand on another
path. Here is another example of this sort. Given (0, 1), consider the evaluation of
Z
(16.24)
B=
0
x
dx.
1 + x2
lim
R
R
z
dz = B e2i B.
1 + z2
X
z
z
dz
=
2i
Res
p
1 + z2
1 + z2
p=i
= i ei(1)/2 + e3i(1)/2 ,
so
(16.27)
B = i
ei(1)/2 + e3i(1)/2
cos (1 )/2
=
.
2i
1e
sin (1 )
133
While all of the examples above involved integrals over infinite intervals, one can also
use residue calculus to evaluate integrals of the form
Z
(16.28)
R(cos , sin ) d,
0
when R(u, v) is a rational function of its arguments. Indeed, if we consider the curve
() = ei , 0 2, we see that (16.28) is equal to
Z
1 z
1 dz
z
+ ,
,
(16.29)
R
2 2z 2i 2iz iz
Exercises
1. Use residue calculus to evaluate the following definite integrals.
Z
x2
(a)
dx,
6
1 + x
Z
cos x
dx,
1 + x2
x1/3
dx,
1 + x3
(b)
0
Z
(c)
0
(d)
0
sin2 x
dx.
2 + cos x
Z
0
xa
dx,
1 + xn
134
3. If
n
k
d
,
1 2r cos + r2
f
f (p)
= 0 .
g
g (p)
Resp
f (z)
= f 0 (p).
(z p)2
Resp
f (z)
1
=
f (k1) (p).
k
(z p)
(k 1)!
Hint. Use (5.11), in concert with (16.2). See how this generalizes (16.5) and (16.30).
8. Compute
for R.
eix
dx,
1 + x4
135
(17.2)
1 X
2i j=1
f 0 (z)
dz,
f (z)
Dj
1
mj =
2i
f 0 (z)
dz.
f (z)
Dj
gj0 (z)
f 0 (z)
mj
=
+
.
f (z)
z pj
gj (z)
136
Having Proposition 17.1, we now want to interpret (17.1) in terms of winding numbers.
Denote the connected components of by Cj (with proper orientations). Say Cj is
parametrized by j : S 1 C. Then
f j : S 1 C \ 0
(17.7)
Cj
Note that the right side of (17.1) is equal to the sum of the terms (17.8). As z runs over
j , we write z in polar coordinates, z = rei . Hence
(17.9)
dz = ei dr + irei d,
so
dr
dz
=
+ i d.
z
r
R
Noting that dr/r = d(log r) and that j d(log r) = 0, we have
Z
Z
1
1
dz
(17.11)
=
d.
2i
z
2
(17.10)
This theta integral is not necessarily zero, since is not single valued on C \ 0. Rather we
have, for any closed curve in C \ 0,
Z
1
(17.12)
d = n(, 0),
2
the winding number of about 0. Certainly the amount by which changes over such a
curve is an integral multiple of 2, so n(, 0) is an integer.
We can provide a formula for d in terms of single valued functions, as follows. We start
with
(17.13)
dz
dx + idy
(x iy)(dx + idy)
=
=
z
x + iy
x2 + y 2
x dx + y dy
x dy y dx
=
+i 2
,
2
2
x +y
x + y2
d =
x dy y dx
.
x2 + y 2
137
1
2i
f 0 (z)
dz = n(j , 0),
f (z)
j = f (Cj ).
Cj
(f, ) =
n(j , 0),
j = f (Cj ).
That is, the total number of zeros of f in , counting multiplicity, is equal to the sum of
the winding numbers of f (Cj ) about 0.
This result is called the argument principle. It is of frequent use, since the right side
of (17.18) is often more readily calculable directly than the left side. In evaluating this
sum, take care as to the orientation of each component Cj , as that affects the sign of the
winding number. We mention without further ado that the smoothness hypothesis on
can be relaxed via limiting arguments.
The following useful corollary of Proposition 17.4 is known as Rouches theorem or the
dogwalking theorem.
Proposition 17.5. Let f, g C 1 () be holomorphic in , and nowhere zero on .
Assume
(17.19)
z .
Then
(17.20)
(f, ) = (g, ).
138
Proof. The hypothesis (17.19) implies that f and g are smoothly homotopic as maps from
to C \ 0, e.g., via the homotopy
f (z) = f (z) [f (z) g(z)],
0 1.
Hence, by Proposition 17.2, f Cj and gCj have the same winding numbers about 0, for
each boundary component Cj .
As an example of how this applies, we can give another proof of the fundamental theorem
of algebra. Consider
(17.21)
f (z) = z n ,
g(z) = z n + an1 z n1 + + a0 .
for z = R.
f (z) = f (p) +
an (z p)n ,
n=k
where we pick ak to be the first nonzero coefficient. It follows that there is a disk D (p)
such that = f D (p) is bounded away from q, and the winding number of this curve
about q is equal to k (which is 1). It follows that there exists > 0 such that whenever
q 0 q < then the winding number of about q 0 is equal to k, so f (z) q 0 has zeros
in D (p). This shows that {q 0 C : q 0 q < } is contained in the range of f , so the
proposition is proven.
The argument principle also holds for meromorphic functions. We have the following
result.
Proposition 17.7. Assume f is meromorphic on a bounded domain , and C 1 on a
neighborhood of . Then the number of zeros of f minus the number of poles of f
(counting multiplicity) in is equal to the sum of the winding numbers of f (Cj ) about 0,
where Cj are the connected components of .
Proof. The identity (17.1), with (f, ) equal to zeros minus poles, follows by the same
reasoning as used in the proof of Proposition 17.1, and the interpretation of the right side
of (17.1) in terms of winding numbers follows as before.
Another application of Proposition 17.1 yields the following result, known as Hurwitz
theorem.
139
(17.25)
Z
D
fn0 (z)
dz = 0,
fn (z)
n.
1
(f, D) =
2i
f 0 (z)
dz = 0,
f (z)
Exercises
1. Let f (z) = z 3 + iz 2 2iz + 2. Compute the change in the argument of f (z) as z varies
along:
a) the real axis from 0 to ,
b) the imaginary axis from 0 to ,
c) the quarter circle z = Rei , where R is large and 0 /2.
Use this information to determine the number of zeros of f in the first quadrant.
2. Prove that for any > 0 the function
1
+ sin z
z+i
has infinitely many zeros in the strip  Im z < .
Hint. Rouches theorem.
3. Suppose f : C is holomorphic and onetoone. Show that f 0 (p) 6= 0 for all p ,
using the argument principle.
Hint. Compare the proof of Proposition 17.6, the open mapping theorem.
4. Make use of Exercise 7 in 5 to produce another proof of the open mapping theorem.
140
5. Let C be open and connected and assume gn : C are holomorphic and each
is oneto one (we say univalent). Assume gn g locally uniformly on . Show that g is
either univalent or constant.
Hint. Pick arbitrary b and consider fn (z) = gn (z) gn (b).
6. Let D be a disk in C. Assume f C 1 (D) is holomorphic in D. Show that f (D)
cannot be a figure 8.
7. In the setting of Proposition 17.1, assume S C is connected and S f () = . Show
that
(f p, ) is independent of p S.
Hint. Use (17.1) to show that (p) = (f p, ) gives a continuous function : S Z.
8. Let > 1. Show that zez = 1 for exactly one z D = {z C : z < 1}.
Hint. With f (z) = zez , show that
z = 1 = f (z) > 1.
Compare the number of solutions to f (z) = 0. Use either Exercise 17, with S = D, or
Rouches theorem. with f (z) = zez and g(z) = zez 1.
In Exercises 912, we consider
(z) = ze
: D C,
: D C.
p ( p, D) = 0.
Hint. For the first case, take p = 0. For the second, let p .
11. Deduce that : D + is onetoone and onto.
12. Recalling Exercise 8, show that {z C : z < 1/e} + .
141
(z) =
et tz1 dt,
Re z > 0.
Here we develop further properties of this special function, beginning with the following
crucial identity:
Z
(z + 1) =
et tz dt
0
Z
d t z
(18.3)
(e ) t dt
=
dt
0
= z (z),
for Re z > 0, where we use integration by parts. The definition (18.2) clearly gives
(18.4)
(1) = 1,
While (z) is defined in (18.2) for Re z > 0, note that the left side of (18.3) is well defined
for Re z > 1, so this identity extends (z) to be meromorphic on {z : Re z > 1}, with
a simple pole at z = 0. Iterating this argument, we extend (z) to be meromorphic on C,
with simple poles at z = 0, 1, 2, . . . . Having such a meromorphic continuation of (z),
we establish the following identity.
Proposition 18.1. For z C \ Z we have
(18.6)
(z)(1 z) =
.
sin z
Proof. It suffices to establish this identity for 0 < Re z < 1. In that case we have
Z Z
(z)(1 z) =
e(s+t) sz tz1 ds dt
Z0 Z0
(18.7)
=
eu v z1 (1 + v)1 du dv
Z0 0
=
(1 + v)1 v z1 dv,
0
142
where we have used the change of variables u = s + t, v = t/s. With v = ex , the last
integral is equal to
Z
(18.8)
(1 + ex )1 exz dx,
which is holomorphic on 0 < Re z < 1. We want to show that this is equal to the right
side of (18.6) on this strip. It suffices to prove identity on the line z = 1/2 + i, R.
Then (18.8) is equal to the Fourier integral
Z
x 1 ix
(18.9)
2 cosh
e dx.
2
,
cosh
(18.10)
and since
(18.11)
sin ( 12 + i)
,
cosh
t n
t2
0 et 1
et ,
n
n
0 t n,
(18.13)
t n
t
log 1
= n log 1
= t X,
n
n
1 t 2
t2 1 1 t
X=
+
+
+ .
n 2 3n 4 n
t n
et 1
= (1 eX )et Xet ,
n
143
using the estimate x (1 ex ) 0 for x 0 (as above). It is clear from (18.13) that
X t2 /n if t n/2. On the other hand, if t n/2 and n 4 we have t2 /n 1 and hence
et (t2 /n)et .
We use (18.12) to obtain, for Re z > 0,
Z n
t n z1
(z) = lim
1
t
dt
n 0
n
Z 1
z
= lim n
(1 s)n sz1 ds.
n
n(n 1) 1
(z) = lim n
n
z(z + 1) (z + n 1)
sz+n1 ds,
(18.15)
1 2n
,
z(z + 1) (z + n)
Using the identity (18.3), analytically continuing (z), we have (18.15) for all z C
other than 0, 1, 2, . . . . In more detail, we have
(z) =
(z + 1)
1
1 2n
= lim nz+1
,
n
z
z (z + 1)(z + 2) (z + 1 + n)
,
z(z + 1) (z + n + 1)
n+1
nz
z 1
z 1
z 1
1
(18.16)
(z) = lim n z (1 + z)
1+
1 +
.
n
2
n
To work on this formula, we define Eulers constant:
1
1
(18.17)
= lim 1 + + + log n .
n
2
n
Then (18.16) is equivalent to
z 1
z 1
z z(1+1/2++1/n) 1
1
1+
(18.18)
(z) = lim e
e
z (1 + z)
1 +
,
n
2
n
which leads to the following Euler product expansion.
144
(18.19)
We can combine (18.6) and (18.19) to produce a product expansion for sin z. In fact,
it follows from (18.19) that the entire function 1/(z)(z) has the product expansion
Y
z2
1
2
= z
1 2 .
(z)(z)
n
n=1
(18.20)
sin z = z
Y
n=1
z2
1 2 .
n
= .
2
Another way to obtain (18.22) is the following. A change of variable gives
Z
Z
1 t 1/2
1 1
x2
(18.23)
e
dx =
e t
dt =
.
2 0
2 2
0
It follows from (10.6) that the left side of (18.23) is equal to /2, so we again obtain
(18.22). Note that application of (18.3) then gives, for each integer k 1,
1
1
1
3
(18.24)
k+
= 1/2 k
k
.
2
2
2
2
One can calculate the area An1 of the unit sphere S n1 Rn by relating Gaussian
integrals to the Gamma function. To see this, note that the argument giving (10.6) yields
Z
Z
n
2
x2
e
dx =
ex dx = n/2 .
(18.25)
Rn
On the other hand, using spherical polar coordinates to compute the left side of (18.24)
gives
Z
Z
2
x2
e
dx = An1
er rn1 dr
(18.26)
Rn
1
= An1
2
Z
0
et tn/21 dt,
145
An1 =
2 n/2
.
(n/2)
Exercises
1. Use the product expansion (18.19) to prove that
d 0 (z) X
1
=
.
2
dz (z)
(z
+
n)
n=0
(18.28)
Hint. Go from (18.19) to
X
1
z zi
log
= log z + z +
log 1 +
,
(z)
n
n
n=1
d 0 (z)
d2
= 2 log (z).
dz (z)
dz
2. Let
1
1
+ + log(n + 1).
2
n
Show that n % and that 0 < n < 1. Deduce that = limn n exists, as asserted in
(18.17).
n = 1 +
(Re z > 0)
Re s > 0.
(z + 1) = z(z) = e
Y
z 1 z/n
1+
e ,
n
n=1
z < 1.
146
(18.30)
d
z(z) z=0 = .
dz
log s +
,
s
a =
0
1 et
dt,
t
b =
1
et
dt.
t
B(z, ) =
(z)()
,
(z + )
Re z, Re > 0,
B(z, ) =
Re z, Re > 0.
The identity (18.31) is known as Eulers formula for the beta function.
8. Show that, for any z C, when n 2z, we have
z z/n
e
= 1 + wn
1+
n
with log(1 + wn ) = log(1 + z/n) z/n satisfying
2
log(1 + wn ) z .
n2
147
Show that this estimate implies the convergence of the product on the right side of (18.19),
locally uniformly on C.
More infinite products
9. Show that
Y
(18.33)
n=1
1
2
1 2 = .
4n
cos
z
=
2
Y
z2
1 2 .
n
odd n1
(z 1)2
cos
= (1 z)
1
.
2
2
2
4n
n=1
(18.35)
1+
1
z
1
z
1
+
2n 2n
2n 2n
1
z
z
= 1 2 1
1+
,
4n
2n + 1
2n 1
z
Y
1
cos
=
1 2
2
2 n=1
4n
Y
odd n1
z
z
1
1+
.
n
n
sin z
z
z
z
= cos
cos
cos
.
z
2
4
8
148
1
(18.37)
(2z) = eA(z) (z) z +
,
2
with A(z) holomorphic on C. We seek a formula for A(z). We will be guided by (18.19),
which implies that
Y
2z 2z/n
1
2z
= 2ze
1+
e
,
(2z)
n
n=1
(18.38)
(18.39)
1
(z)(z + 1/2)
1 z (z+1/2) n Y
z z/n
z + 1/2 (z+1/2)/n o
=z z+
e e
1+
e
1+
e
.
2
n
n
n=1
Setting
(18.40)
2z + 2n + 1
2z
1
z + 1/2
=
= 1+
1+
,
1+
n
2n
2n + 1
2n
and
(18.41)
(18.42)
1+
n=1
2z 2z/2n
2z 2z/(2n+1) o
e
1+
e
2n
2n + 1
nY
n=1
1+
1 1/2n o Y 2z[(1/2n)1/(2n+1)]
e
e
.
2n
n=1
Hence
(18.43)
1
e2z
= ze2z e/2
(1 + 2z)e2z (18.42)
(z)(z + 1/2)
2
2z n Y
2z 2z/k o
2z /2 e
= 2ze e
1+
e
4
k
k=1
nY
n=1
1 1/2n o Y 2z[(1/2n)1/(2n+1)]
1+
e
e
.
2n
n=1
149
1
1 /2 Y
1 1/2n
= e
1+
e
,
(1/2)
2
2n
n=1
(18.44)
(18.45)
1
1
e2z Y 2z[(1/2n)1/(2n+1)]
=
e
(z)(z + 1/2)
(1/2)(2z) 2 n=1
e2z
1
,
(1/2)(2z) 2
where
(18.46)
X
1
1
=1
2n 2n + 1
n=1
1 1 1 1
+ +
2 3 4 5
= log 2.
=1
1
(18.47)
.
2
2
,
n
n
n
valid for n = 3, 4, . . . .
(1 + ak ).
k=1
150
Disregarding cases where one or more factors 1+ak vanish, the convergence of
as M amounts to the convergence
(18.51)
lim
N
Y
(1 + ak ) = 1,
QM
k=1 (1+ak )
uniformly in N > M.
k=M
(18.52)
(1 + ak ) = (1 + aM )(1 + aM +1 ) (1 + aN )
k=M
=1+
aj +
aj1 aj2 + + aM aN ,
j1 <j2
aj  +
aj1 aj2  + + aM aN 
(1
+
a
)
k
j
k=M
(18.53)
=
j1 <j2
N
Y
(1 + ak ) 1
k=M
= bM N ,
the last identity defining bM N . Our task is to investigate when bM N 0 as M ,
uniformly in N > M . To do this, we note that
N
Y
log(1 + bM N ) = log
(1 + ak )
k=M
(18.54)
=
N
X
k=M
x 0 = log(1 + x) x,
x
0 x 1 = log(1 + x) .
2
N
N
X
1 X
ak  log(1 + bM N )
ak ,
2
k=M
k=M
151
and hence
(18.57)
ak  < .
Consequently,
X
ak  < =
(18.58)
Y
k=1
(1 + ak ) converges
(1 + ak ) converges.
k=1
ak  <
(1 + ak ) 6= 0.
k=1
We can replace the sequence (ak ) of complex numbers by a sequence (fk ) of holomorphic
functions, and deduce from the estimates above the following.
Proposition 18.6. Let fk : C be holomorphic. If
X
(18.60)
(18.61)
(1 + fk (z)) = g(z),
k=1
nY
k=1
oY
Y
(1 + fk (z))
(1 + gk (z)) =
(1 + fk (z))(1 + gk (z)).
k=1
k=1
To make contact with the Gamma function, note that the infinite product in (18.19)
has the form (18.61) with
(18.63)
z z/k
e
.
1 + fk (z) = 1 +
k
152
ew = 1 w + R(w),
Hence
1+
(18.65)
z
z z/k
z
z
e
= 1+
1 +R
k
k
k
k
2
z
z
z
R
.
=1 2 + 1
k
k
k
fk (z) =
z z
z2
+
1
R
,
k2
k
k
so
(18.67)
which yields (18.60).
z 2
fk (z) C for k z,
k
153
(s) =
ns ,
Re s > 1.
n=1
2
,
6
(2) =
(4) =
4
.
90
This function is of great interest in number theory, due to the following result.
Proposition 19.1. Let {pj : j 1} = {2, 3, 5, 7, 11, . . . } denote the set of prime numbers
in N. Then, for Re s > 1,
(19.3)
(s) =
1
(1 ps
.
j )
j=1
(19.4)
2s
(1 + ps
+ p3s
+ )
j + pj
j
j=1
=1+
ps
j +
(pj1 pj2 )s +
j1 j2
j1 j2 j3
That this is identical to the right side of (19.1) follows from the fundamental theorem of
arithmetic, which says that each integer n 2 has a unique factorization into a product
of primes.
From (19.1) we see that
(19.5)
s & 1 = (s) % +.
Hence
(19.6)
(1 p1
j ) = 0.
j=1
154
X
1
= ,
p
j=1 j
(19.7)
X
1
s  < ,
p
j
j=1
(19.8)
for Re s > 1.
(s) 6= 0,
for Re s > 1.
g(t) =
en
n=1
Indeed, we have
Z
(19.11)
g(t)t
s1
dt =
2s s
n=1
= (2s)
et ts1 dt
(s).
This gives rise to further useful identities, via the Jacobi identity (14.43), i.e.,
r
X
1 X k2 /t
`2 t
(19.12)
e
=
e
,
t
`=
k=
which implies
(19.13)
1
1 1 1/2
1/2
g(t) = + t
+t
g
.
2 2
t
(s) =
g(t)ts/21 dt
2
0
(19.14)
Z 1
Z
s/21
=
g(t)t
dt +
g(t)ts/21 dt.
0
155
Into the integral over [0, 1] we substitute the right side of (19.13) for g(t), to obtain
Z 1
s
1 1
s/2
(s) =
+ t1/2 ts/21 dt
2
2 2
0
(19.15)
Z 1
Z
g(t1 )ts/23/2 dt +
+
0
g(t)ts/21 dt.
We evaluate the first integral on the right, and replace t by 1/t in the second integral, to
obtain, for Re s > 1,
Z
s
s/2
1
1
s/2
(19.16)
(s) =
+
t + t(1s)/2 g(t)t1 dt.
2
s1 s
1
Note that g(t) Cet for t [1, ), so the integral on the right side of (19.16) defines
an entire function of s. Since 1/(s/2) is entire, with simple zeros at s = 0, 2, 4, . . . , as
seen in 18, this implies that (s) is continued as a meromorphic function on C, with one
simple pole, at s = 1. This finishes the proof of Proposition 19.2.
The formula (19.16) does more than establish the meromorphic continuation of the zeta
function. Note that the right side of (19.16) is invariant under replacing s by 1 s. Thus
we have an identity known as Riemanns functional equation:
s
1 s
(19.17)
s/2 (s) =
(1s)/2 (1 s).
2
2
The meromorphic continuation of (s) can be used to obtain facts about the set of
primes deeper than (19.7). It is possible to strengthen (19.9) to
(19.18)
(s) 6= 0
for Re s 1.
lim
pj
= 1.
j log j
A proof can be found in [BN], and in [Ed]. Of course, (19.19) is much more precise than
(19.7).
There has been a great deal of work on determining where (s) can vanish. By (19.16),
it must vanish at all the poles of (s/2), other than s = 0, i.e.,
(19.20)
(s) = 0 on {2, 4, 6, . . . }.
These are known as the trivial zeros of (s). It follows from (19.18) and the functional
equation (19.16) that all the other zeros of (s) are contained in the critical strip
(19.21)
Concerning where in these zeros can be, there is the following famous conjecture.
156
The Riemann hypothesis. All zeros in of (s) lie on the critical line
n1
(19.22)
o
+ i : R .
Many zeros of (s) have been computed and shown to lie on this line, but after over a
century, a proof (or refutation) of the Riemann hypothesis eludes the mathematics community. The reader can consult [Ed] for more on the zeta function.
Exercises
1. Use the functional equation (19.17) together with (18.6) and the Legendre duplication
formula (18.47) to show that
s
(1 s) = 21s s cos
(s)(s).
2
2. Sum the identity
Z
s
(s)n
ent ts1 dt
ts1
dt =
et 1
Z
0
ts1
dt +
et 1
Z
1
ts1
dt = A(s) + B(s).
et 1
et
to show that
1
0
1
1
= + a0 + a1 t + a2 t2 +
1
t
ts1
1
a0
a1
dt =
+
+
+
t
e 1
s1
s
s+1
provides a meromorphic continuation of A(s), with poles at {1, 0, 1, . . . }. Use this to give
a second proof that (s) has a meromorphic continuation with one simple pole, at s = 1.
157
X
d(n)
(s) =
,
ns
n=1
(b)
X
(n)
(s)(s 1) =
,
ns
n=1
(c)
(s 1) X (n)
=
,
s
(s)
n
n=1
(d)
X
1
(n)
=
,
(s) n=1 ns
(e)
where
X
0 (s)
(n)
=
,
s
(s)
n
n=1
d(n) = # divisors of n,
(n) = sum of divisors of n,
(n) = # positive integers n, relatively prime to n,
(n) = (1)# prime factors , if n is squarefree, 0 otherwise,
(n) = log p if n = pm for some prime p, 0 otherwise.
158
fU = 1 f U ,
q
q
(20.1)
exp : C C \ {0},
Sq : C \ {0} C \ {0},
159
f (z) = eg(z) ,
f (z) = h(z)2 .
p
f (z), over U .
Exercises
1. As in Corollary 20.4, suppose U C is simply connected and f : U C \ {0} is
holomorphic.
(a) Show that f 0 /f : U C is holomorphic.
(b) Pick p U and define
Z
(z) =
p
f 0 ()
d.
f ()
Show this is independent of the choice of path in U from p to z, and it gives a holomorphic
function : U C.
(c) Use to give another proof of Corollary 20.4 (particularly the existence of g(z) in
(20.3)).
160
If the identity of the pair (, S) is understood, we omit the phrase with respect to (, S).
The main case of interest to us here is where C is an open set and S = C. However,
in later sections the case S = C {} S 2 will also be of interest.
A major technique to identify normal families is the following result, known as the
ArzelaAscoli theorem.
Proposition 21.1. Let X and Y be compact metric spaces and fix a modulus of continuity
(). Then
(21.2)
x, y X, f F ,
161
f  Aj on Dj ,
f F,
f0 
Aj
on Dj ,
j
f F ,
hence
(21.6)
f (z) f (w)
Aj
z w,
j
z, w Dj , f F.
This equicontinuity on each Dj makes Proposition 21.2 applicable. This establishes one
implication in Proposition 21.3, and the reverse implication is easy.
Exercises
1. Show whether each of the following families is or is not normal (with respect to (, C)).
(a) {n1 cos nz : n = 1, 2, 3, . . . },
162
Df (p) = (p)R(p).
Now det Df (p) > 0 det R(p) = +1 and det Df (p) < 0 det R(p) = 1. In the
former case, R(p) has the form
(22.2)
R(p) =
cos
sin
sin
cos
of a rotation matrix, and we see that Df (p) commutes with J, given by (1.39). In the
latter case, R(p) has the form
(22.3)
R(p) =
cos
sin
sin
cos
(22.4)
C=
1
0
0
1
az + b
LA (z) =
,
cz + d
A=
a b
c d
163
LA : C {} C {}
by setting
(22.7)
LA (d/c) = ,
if c 6= 0,
and
(22.8)
LA () =
a
if c 6= 0,
c
if c = 0.
LA LB = LAB .
In particular LA is bijective in (22.6), with inverse LA1 . If we give C {} its natural topology as the onepoint compactification of C, we have LA a homeomorphism in
(22.6). Later we will give C {} the structure of a Riemann surface and see that LA is
biholomorphic on this surface.
Note that LsA = LA for any nonzero s C. In particular, LA = LA1 for some A1
of determinant 1; we say A1 Sl(2, C). Given Aj Sl(2, C), LA1 = LA2 if and only
if A2 = A1 . In other words, the group of linear fractional transformations (22.5) is
isomorphic to
(22.10)
Note that if a, b, c, d are all real, then LA in (22.5) preserves R {}. In this case we
have A Gl(2, R). We still have LsA = LA for all nonzero s, but we need s R to get
sA Gl(2, R). We can write LA = LA1 for A1 Sl(2, R) if A Gl(2, R) and det A > 0.
We can also verify that
(22.11)
A Sl(2, R) = LA : U U,
164
We now single out for attention the following linear fractional transformation:
zi
1 i
(22.13)
(z) =
, (z) = LA0 (z), A0 =
.
1 i
z+i
It is clear that
(22.14)
: R {} S 1 = D,
: U D,
where D = {z : z < 1} is the unit disk. Note that conjugating the Sl(2, R) action on U
by yields the mappings
MA = LA0 AA1 : D D.
(22.15)
1 (a + d)i b + c (a d)i + b + c
1
(22.16)
A0 AA0 =
.
2i (a d)i b c (a + d)i + b c
It follows that
(22.17)
A0 Sl(2, R) A1
0
n
=
o
Gl(2, C) : 2 2 = 1 .
z +
,
(22.18)
LB : D D, B =
, LB (z) =
z +
2 2 = 1.
The group described on the right side of (22.17) is denoted SU (1, 1). Note that for such
B as in (22.18),
ei +
+ ei
LB (ei ) = i
= ei
,
e +
+ ei
and in the last fraction the numerator is the complex conjugate of the denominator. This
directly implies the result LB : D D for such B.
We have the following important transitivity properties.
Proposition 22.2. The group SU (1, 1) acts transitively on D, via (22.18). Hence Sl(2, R)
acts transitively on U via (22.5).
Proof. Given p D, we can pick , C such that 2 2 = 1 and / = p. Then
LB (0) = p. This together with LB1 L1
B2 = LB1 B21 is enough to establish transitivity of
SU (1, 1) on D. As for transitivity of Sl(2, R) on U, one can use the conjugation (22.17),
or the following direct argument. Given p = a + ib U (a R, b > 0), Lp (z) = bz + a =
(b1/2 z + b1/2 a)/b1/2 maps i to p.
The following converse to Proposition 22.2 is useful.
165
%1 z
Hence the same can be said for h , and then a maximum principle argument yields
g(z)/z 1 on D and also z/g(z) 1 on D; hence g(z)/z 1 on D. This implies
g(z)/z = c, a constant, and that c = 1, as asserted.
c 0
To proceed, we have g = LB2 with B2 = a
, and we can take a = c1/2 , to
0 1
obtain B2 SU (1, 1). We have f = LB 1 B2 , and the proof is complete.
1
We single out some building blocks for the group of linear fractional transformations,
namely (with a 6= 0)
1
.
z
We call these respectively (complex) dilations, translations, and inversion about the unit
circle {z : z = 1}. These have the form (22.5), with A given respectively by
a 0
1 b
0 1
(22.20)
,
,
.
0 1
0 1
1 0
(22.19)
a (z) = az,
b (z) = z + b,
(z) =
D = p r 1/r p .
The reader can work out the explicit form of this linear fractional transformation. Note
that D leaves D invariant and interchanges p and .
Recall that the linear fractional transformation in (22.13) was seen to map R {}
to S 1 . Similarly its inverse, given by i(z) with
(22.22)
(z) =
z+1
,
z1
(ei ) =
ei + 1
i
=
.
i
e 1
tan /2
These are special cases of an important general property of linear fractional transformations. To state it, let us say that an extended line is a set ` {}, where ` is a line in
C.
166
L(ei ) =
(22.24)
L(z) =
az + b
,
z
L(ei ) = a + bei .
167
(22.26)
The simple connectivity of guarantees the existence of such g, as shown in 20. Now
e
g is onetoone on and it maps diffeomorphically onto a simply connected region
having the property that
e = z
e
(22.27)
z
/ .
e and if we compose g with inversion across D
It follows that there is a disk D C \ ,
we obtain such a desired holomorphic diffeomorphism.
Exercises
1. Find conformal mappings of each of the following regions onto the unit disk. In each
case, you can express the map as a composition of various conformal maps.
(a) = {z = x + iy : y > 0, z > 1}.
(b) = C \ (, 1] [1, ) .
(c) = {z : z + 1/2 < 1} {z : z 1/2 < 1}.
2. Consider the quarterplane
= {z = x + iy : x > 0, y > 0}.
Find a conformal map of onto itself such that
(1 + i) = 2 + i.
3. Let f : O be a conformal diffeomorphism. Show that if u : O R is harmonic, so
is u f : R.
4. Write out the details to establish Claims 34 in the proof of Proposition 22.4.
5. Reconsider Exercise 1b) of 21, mapping the region U defined there conformally onto
the unit disk.
6. Given q D = {z : z < 1}, define
zq
.
1 qz
Show that q : D D and q (q) = 0. Write q in the form (22.18). Relate this to the
proof of Proposition 22.2.
(22.28)
q (z) =
168
f : D
g 0 (p)
1
,
R
g F,
so we can set
(23.3)
f : D
is holomorphic and satisfies f 0 (p) = A. We claim this function provides the holomorphic
diffeomorphism (23.1). There are two parts to showing this, treated in the next two
lemmas.
Lemma 23.2. In (23.4), f is onetoone.
Proof. Suppose there exist distinct z1 , z2 such that f (z1 ) = f (z2 ) = w D. Let
O be a smoothly bounded region such that z1 , z2 O and such that f () is disjoint
from w, where = O.
By the argument principle (Proposition 17.4), f () winds (at least) twice about w. But
each g () winds only once about w. Since g f uniformly on , this is a contradiction.
169
F : D.
G(z) =
G : D.
G0 (p) =
F 0 (p)
1 + q
= p A > A.
2
1 F (p)
2 q
Hence G F and G0 (p) > A, a contradiction. This proves Lemma 23.3, and hence the
Riemann mapping theorem.
Exercises
1. Note that F (z) in (23.5) is given by
q
(23.10)
F (z) =
q f (z) ,
G(z) =
F 0 (p)
F (p) F (z) .
0
F (p)
h (0) = 0,
h0 (0) > 0,
h (D) D ,
1.
170
h (z) z
locally uniformly on D.
Hint. Use a normal families argument, and show that any limit hk g must have the
property that g(D) = D, and conclude that g(z) = z. (The argument principle may be
useful.)
3. Suppose is a bounded, simply connected domain, p , and f : D are univalent
holomorphic maps satisfying
(23.15)
f (p) = 0,
f0 (p) > 0,
f () D ,
1.
f2 (z) =
F10 (p)
F1 (p) F1 (z) ,
0
F1 (p)
F1 (z) =
q1 f1 (z) .
Evaluate A2 = f20 (p). Take q2 D \ f2 () with minimal absolute value and use this to
construct f3 . Continue, obtaining f4 , f5 , . . . . Show that at least one of the following holds:
(23.17)
f0 (p) A,
or f () D ,
1,
171
z = f (z) 1.
z O+ , z L = O R = f (z ) 1.
It follows from the form of the Schwarz reflection principle given in 13 that g = f O+
has a holomorphic extension G : O C, and G : L D. Say U = G(O), as in Fig. 24.1.
Note that U is invariant under z 7 z 1 .
Then we have a holomorphic map
(24.3)
F = G 1 : V U.
172
(24.4)
Z
f 0 (z)2 ds.
r
R
2
f 0 (z)2 dx dy = 2 Area f (, R) 2 2 ,
(24.5)
log 2
(,R)
Exercises
1. Suppose f : D is a holomorphic diffeomorphism, extending to a homeomorphism
f : D. Let g C() be given. Show that the Dirichlet problem
(24.6)
u = 0 in , u = g
has a unique solution u C 2 () C(), given by u = v f , where
(24.7)
v = 0 in D, v D = g f 1 D ,
the solution to (24.7) having been given in 13.
173
(1) = i,
(0) = 1,
(1) = i.
(z) = i
(z) 1
.
(z) + 1
4. Show that (ip) = 1 (so (z) as z ip). Show that extends continuously to
R \ {ip} C and
(24.10)
(1) = 1,
(0) = 0,
(1) = 1.
5. Show that you can apply the Schwarz reflection principle repeatedly and extend to a
meromorphic function on C, with simple poles at ip + 4k + 2i`p, k, ` Z. Show that
(24.11)
(z + 4) = (z + 2ip) = (z).
Hint. To treat reflection across the top boundary of R, apply Schwarz reflection to 1/.
Remark. We say is doubly periodic, with periods 4 and 2ip.
6. Say (1 + ip) = r. Show that r > 0, that (1 + ip) = r, and that
(24.12)
ip
2
z =
r
.
(z)
(1) = 1,
(0) = 0,
3i. Let : T U
(1) = 1.
174
(i 3) = .
Let
# + z = + e2i/3 z .
3
3
# (1) = ,
# (i 3) = 1.
# (z) = (z),
(1) = ,
so
(z) =
() = 1,
z+3
.
z1
9. Let X and Y be compact metric spaces, and assume f : X Y is continuous, onetoone, and onto. Show that f is a homeomorphism, i.e., f 1 : Y X is continuous.
Hint. You are to show that if f (xj ) = yj y, then xj f 1 (y). Now since X is compact,
every subsequence of (xj ) has a further subsequence that converges to some point in X...
175
: D C \ {0, 1}.
The proof starts with an examination of the following domain . It is the subdomain
of the unit disk D whose boundary consists of three circular arcs, intersecting D at right
angles, at the points {1, e2i/3 , e2i/3 }. See Fig. 25.1. If we denote by
(25.2)
: D U = {z C : Im z > 0}
the linear fractional transformation of D onto U with the property that (1) = 0, (e2i/3 )
e = () is pictured in Fig. 25.2.
= 1, and (e2i/3 ) = , the image
The Riemann mapping theorem guarantees that there is a holomorphic diffeomorphism
(25.3)
: D,
e U,
= 1 :
: U C \ {0, 1}.
176
Exercises
1. Show that the map : D U in (25.2) is given by
(z) =
z1
,
z 2
= e2i/3 .
Show that
() = 1,
( 2 ) =
1
.
2
1 z
+z = + ,
2
2
4
1
c
1
(x + iy) = x + iy,
+ z = + z.
2
2
z =
(25.6)
1
,
z
(z ) = (z),
(z ) = (z),
(z + 2) = (z).
3. Show that
(25.8)
(z c ) = (z)c .
(z ) = (z) .
1
1
=
.
z
(z)
177
6. Show that (z )c = z + 1 and (z)c = 1 z. Deduce from (25.7) and (25.8) that
(25.11)
(z + 1) = 1 (z).
F01 (w) = 1 w,
F0 (w) =
1
,
w
are holomorphic automorphisms of S 2 that leave C \ {0, 1} invariant, F01 fixes and
switches 0 and 1, while F0 fixes 1 and switches 0 and . Show that these maps generate
a group G of order 6, of automorphisms of S 2 and of C \ {0, 1}, that permutes {0, 1, }
and is isomorphic to the permutation group S3 on three objects. Show that
(25.13)
w
w1
is the element of G that fixes 0 and switches 1 and . Show that the rest of the elements
of G consist of the identity map, w 7 w, and the following two maps:
1
,
1w
w1
.
F10 (w) = F01 F0 (w) =
w
F01 (w) = F0 F01 (w) =
(25.14)
8. Show that the transformations in G listed in (25.12)(25.14) have the following fixed
points.
Element
F0
F01
F1
F01
F10
Fixed points
1, 1 = A1
, 12 = A2
0, 2 = A3
ei/3 = B
ei/3 = B
See Figure 25.3. Show that the elements of G permute {A1 , A2 , A3 } and also permute
{B+ , B }.
9. We claim there is a holomorphic map
(25.15)
H : S 2 S 2 ,
178
satisfying
(25.16)
F G,
such that
(25.17)
(25.18)
(25.19)
H(w) = C
(w ei/3 )3 (w ei/3 )3
(w2 w + 1)3
=
C
,
w2 (w 1)2
w2 (w 1)2
C=
4
,
27
so
(25.22)
H(w) =
4 (w2 w + 1)3
.
27 w2 (w 1)2
Verify that
(25.23)
1
w
= H(w),
1
H
= H (z + 1) = H (z),
z
z U.
For more on these maps, see the exercises at the end of 34. For a different approach, see
the last two sections in Chapter 7 of [Ahl].
179
(26.1)
S(x1 , x2 , x3 ) = (1 x3 )1 (x1 , x2 ).
S 1 (x, y) =
1
(2x, 2y, r2 1),
1 + r2
r 2 = x2 + y 2 .
DF (q) : R2 Tp S 2 R3 .
See Appendix C for more on this. Conformality is equivalent to the statement that there
is a positive function (p) such that, for all v, w R2 ,
(26.5)
or in other words,
(26.6)
1 0
0 1
1 x2 + y 2
2xy
2
2xy
(26.7)
DF (x, y) =
1 + x2 y 2 ,
2
2
(1 + r )
2x
2y
and hence
(26.8)
4
DF (x, y) DF (x, y) =
(1 + r2 )2
t
1
0
0
1
180
S : S 2 \ {e3 } R2 .
S (x1 , x2 , x3 ) = (1 + x3 )1 (x1 , x2 ).
S S 1 (x, y) =
1
(x, y).
r2
S S 1 (z) =
1
z
= .
2
z
z
181
(26.13)
T = C/
Exercises
1. Give an example of a family F of holomorphic functions C with the following two
properties:
(a) F is normal with respect to (, S 2 ).
(b) {f 0 : f F} is not normal, with respect to (, C).
Compare Exercise 2 of 21. See also Exercise 11 below.
2. Given C open, let
F = {f : C : Re f > 0 on , f holomorphic}.
Show that F is normal with respect to (, S 2 ). Is F normal with respect to (, C)?
182
p(z)
.
q(z)
Show that R(z) has a unique continuous extension R : S 2 S 2 , and this is holomorphic.
Exercises 79 deal with holomorphic maps F : S 2 S 2 . Assume F is not constant.
7. Show that there are only finitely many pj S 2 such that DF (pj ) : Tpj S 2 Tqj S 2 is
singular (hence zero), where qj = F (pj ). The points qj are called critical values of F .
8. Suppose is not a critical value of F and that F 1 () = {, p1 , . . . , pk }. Show that
f (z) = F (z)(z p1 ) (z pk ) : C C,
and f (z) as z . Deduce that f (z) is a polynomial in z. (Cf. Proposition
11.4.)
9. Show that every holomorphic map F : S 2 S 2 is of the form treated in Exercise 6
(except for the constant map F ).
Hint. Compose with linear fractional transformations and transform F to a map satisfying
the conditions of Exercise 8.
10. Given a holomorphic map f : C, set
(26.13)
g = S 1 f : S 2 .
Dg(z) : R2 Tp S 2 .
183
Dg(z)t Dg(z) = 4
f 0 (z) 2
I,
1 + f (z)2
(26.16)
f 0 (z)
1 + f (z)2
{f # (z) : f F, z K} is bounded.
(26.18)
where = {4k + 2i`p : k, ` Z}.
S : S 2 C {},
bijective,
with S(e3 ) = .
13. For R, define : C C by (z) = ei z. Also set () = . Show that
cos sin
.
(26.20)
R = S 1 S = R = sin
cos
1
14. Let R : S 2 S 2 be a conformal diffeomorphism with the properties
R(e3 ) = e3 ,
R : E E,
184
f (z) =
(cos )z sin
,
(sin )z + cos
f : C {} C {}.
Set
= S 1 f S,
(26.22)
: S 2 S 2 .
e E,
e
: E
f : R {} R {},
f () =
cos
.
sin
cos 2
sin 2
.
(26.23)
=
1
sin 2
cos 2
Hint. Translate the result of Exercise 14 to this setting.
To A G`(2, C) we associate the linear fractional transformation LA as in (22.5),
(26.24)
az + b
LA (z) =
,
cz + d
A=
a b
c d
LA : C {} C {}.
As also define
(26.25)
A = S 1 LA S,
A : S 2 S 2 .
185
(q) = q 0 ,
(r) = r0 .
186
g (p ) = D1 .
g g : U D.
g (K) D1 ,
187
f 0,
f 1,
or f : U S 2 \ {0, 1, }.
f ,
E =
{EO : O neighborhood of }.
E = {0, },
JSq .
188
8. Suppose E consists of one point. Show that R is conjugate to a polynomial, i.e., there
exists a linear fractional transformation L such that L1 RL is a polynomial.
Hint. Consider the case E = {}.
9. Suppose E consists of two points. Show that R is conjugate to Pm for some m Z,
where Pm (z) = z m , defined appropriately at 0 and .
Hint. Suppose E = {0, }. Then R either fixes 0 and or interchanges them.
Conclusion. Typically, E = , for JR . Furthermore, if E 6= , then E = E is
independent of JR , and E FR .
10. Show that
(27.8)
R : JR JR
and
R1 (JR ) JR .
(27.9)
Hint. Use Exercise 4. For (27.8), if R() FR , then has a neighborhood O such that
R(O) FR . For (27.9), use R : FR FR .
11. Show that either JR = S 2 or JR has empty interior.
Hint. If JR has a neighborhood O JR , then, by (27.8), Rn (O) JR . Now use
Exercise 5.
12. Show that, if p JR , then
(27.10)
Rk (p) is dense in JR .
k0
Hint. Use Exercises 56, the conclusion following Exercise 9, and Exercise 10.
13. Show that
(27.11)
R(z) = 1
2
= JR = S 2 .
z2
Remark. This could be tough. See [CG], p. 82, for a proof of (27.11), using results not
developed in these exercises.
14. Show that
(27.12)
R(z) = z 2 2 = JR = [2, 2] C.
Hint. Show that R : [2, 2] [2, 2], and that if z0 C \ [2, 2], then Rk (z0 ) as
k .
The next exercise will exploit the following general result.
189
F j (U ) = X.
0jN (U )
F j (p) is dense in X.
j1
Rj (p) is dense in JR .
j1
j 1.
190
rest of JR , and (again) apply Exercise 5 (and Exercise 15) to R, to get a contradiction,
taking into account Exercise 20.
These exercises provide a glimpse at an area known as Complex Dynamics. More material
on this area can be found in Chapter 5 of [Sch] (a brief treatment), and in [CG] and [Mil].
As opposed to (27.7) and (27.12), typically JR has an extremely complicated, fractal
structure, as explained in these references.
191
g : C D,
f = g.
f : D \ {0} C \ {p, q}
is holomorphic, then the singularity at 0 is either a pole or a removable singularity. Equivalently, if f : D \ {0} S 2 is holomorphic and has range in S 2 \ {p, q, r} with p, q, r S 2
distinct, then f extends to f : D S 2 , holomorphic.
Proof. Again there is no loss of generality in taking p = 0, q = 1, and r = . Take
= D \ {0} and define f : C \ {0, 1} by f (z) = f (2 z). By Montels theorem
(Theorem 27.1), {f } is a normal family of maps from to S 2 . In particular, there exists
a subsequence fk converging locally uniformly on :
fk g : S 2 ,
(28.3)
fk (z) A,
z = r,
f (z) A,
z = 2k r.
192
The maximum principle then implies f (z) A for all z close to 0, so 0 is a removable
singularity of f .
Case B. g().
By Corollary 27.2, g on . Consequently, 1/fk 0 uniformly on , and then
analogues of (28.4)(28.5) hold with f replaced by 1/f (recall that f is nowhere vanishing
on D \ {0}), so f has a pole at 0.
Exercises
1. Fix distinct points a, b C, let be the line segment joining a and b, and assume
f : C C \ is holomorphic. Show that f is constant. Show this in an elementary
fashion, not using Picard theorems.
2. Suppose f : C C is a holomorphic function that is not a polynomial. (We say f
is transcendental.) Show that for every w C, with at most one exception, there are
infinitely many z C such that f (z ) = w.
3. Suppose that f is meromorphic on C. Show that f can omit at most two complex
values. Give an example where such an omission occurs.
193
29. Harmonic functions again: Harnack estimates and more Liouville theorems
Here we study further properties of harmonic functions. A key tool will be the explicit
formula for the solution to the Dirichlet problem
(29.1)
u = 0 on D1 (0), uS 1 = f,
given in 13, namely
Z
(29.2)
p(z, )f () d,
u(z) =
0
where
(29.3)
p(z, ) =
1 1 z2
,
2 w z2
w = ei ,
as in (13.24), or equivalently
(29.4)
p(z, ) =
1
ei + z
1
ei
1
Re i
= Re i
.
2
e z
e z
2
p(x, ) =
1 x2
,
An1 x n
1
194
u A = u(z)
1 z
.
1 + z
Proof. The mean value property of harmonic functions implies that f = uD satisfies
Z 2
1
(29.9)
f () d = 1,
2 0
if u A. Given this, (29.8) follows from (29.2)(29.3) together with the estimate
(29.10)
1 z2
1 z
,
w z2
1 + z
(29.11)
1 + 2r + r2 = (1 + r)2 ,
from which (29.10) follows.
By translating and scaling, we deduce that if u is 0 and harmonic on DR (p), then
(29.12)
z p = a [0, R) = u(z)
Ra
u(p).
R+a
This leads to the following extension of the Liouville theorem, Proposition 7.5 (with a
different approach from that suggested in Exercise 1 of 7).
Proposition 29.2. If u is harmonic on all of C (and real valued) and bounded from below,
then u is constant.
Proof. Adding a constant if necessary, we can assume u 0 on C. Pick z0 , z1 C and
set a = z0 z1 . By (29.12),
(29.13)
u(z0 )
Ra
u(z1 ),
R+a
R (a, ),
hence, taking R ,
(29.14)
u(z0 ) u(z1 ),
u(z0 ) = u(z1 ),
195
Proposition 29.3. There exists a number A (0, ) with the following property. Let u
be harmonic in DR (0). Assume
(29.16)
u(0) = 0,
u(z) M
on DR (0).
Then
(29.17)
p DR/2 (0),
u(p) =
inf
DR/2 (0)
u.
v(z)
1
M u(p) on DR/2 (p).
3
Hence
(29.20)
1
Area DR (0)
ZZ
v(z) dx dy
1 1
M u(p) .
16 3
DR (0)
On the other hand, the mean value property for harmonic functions implies that the left
side of (29.20) equals v(0) = M , so we get
(29.21)
M u(p) 48M,
u(z) C0 + C1 zk ,
z C.
u(z) C2 C3 zk ,
z C.
196
u(z) C2 + C3 zk ,
z C.
We aim to show that if u(z) is harmonic on C and satisfies the bound (29.24), then u
must be a polynomial in x and y. For this, it is useful to have estimates on derivatives
xi yj u, which we turn to.
If u C 2 (D) C(D) is harmonic on D = D1 (0), the formula (25.2) yields
Z
(29.25)
xi yj u(z)
pij (z, )f () d,
0
where
pij (z, ) = xi yj p(z, ).
(29.26)
From (29.3) it is apparent that p(z, ) is smooth in z D. We have bounds of the form
(29.27)
z
1
.
2
(29.28)
1
ei
p(z, ) = Re i
,
x
(e z)2
1
iei
p(z, ) = Re i
.
y
(e z)2
Hence
1
1
ei z2
1
,
(1 z)2
x,y p(z, )
(29.29)
z1/2
197
u(z) C0 + C1 zk ,
z C,
u(z) C2 + C3 zk ,
z C.
z
1
, R [1, ).
2
Exercises
1. Suppose f : C C is holomorphic and f (z) AeBz . Show that if f has only finitely
many zeros then f has the form
f (z) = p(z) eaz+b ,
for some polynomial p(z).
Hint. If p(z) has the same zeros as f (z), write f (z)/p(z) = eg(z) and apply Proposition
29.5 to Re g(z).
2. Show that the function ez z (considered in Exercise 10 of 6) has infinitely many
zeros. More generally, show that, for each a C, ez z a has infinitely many zeros.
3. The proof given above of Proposition 29.3 shows that (29.17) holds with A = 47. Can
you show it holds with some smaller A?
198
f1 (z) =
1
(z n)2
n=
1
2
=
.
2
2
(z
n)
sin
z
n=
To see this, note that both sides have the same poles, and their difference g1 (z) is seen to
be an entire function, satisfying g1 (z + 1) = g1 (z). Also it is seen that, for z = x + iy, both
sides of (30.2) tend to 0 as y . This forces g1 0.
A second example is
1
1 X 1
1
f2 (z) = lim
= +
+
m
zn
z
zn n
n=m
m
X
(30.3)
n6=0
1
2z
+
.
2
z n=1 z n2
In this case again we see that the difference g2 (z) is entire. Furthermore, applying d/dz
to both sides of (30.4), we get the two sides of (30.2), so g2 is constant. Looking at the
last term in (30.3), we see that the left side of (30.4) is odd in z; so is the right side; hence
g2 = 0.
As a third example, we consider
m
1
X
(1)n
1 X
1
n
lim
= +
(1)
+
m
zn
z
zn n
n=m
(30.5)
n6=0
1
2z
+
(1)n 2
z n=1
z n2
X
1
z(1 2k)
= 4
.
2
z
[z (2k 1)2 ][z 2 (2k)2 ]
k=1
199
We claim that
1
1 X
1
+
(1)n
+
=
.
z
zn n
sin z
(30.6)
n6=0
In this case we see that their difference g3 (z) is entire and satisfies g3 (z + 2) = g3 (z). Also,
for z = x + iy, both sides of (30.6) tend to 0 as y , so g3 0.
We now use a similar device to construct doubly periodic meromorphic functions, following K. Weierstrass. These functions are also called elliptic functions. Further introductory
material on this topic can be found in [Ahl] and [Hil]. Pick 1 , 2 C, linearly independent
over R, and form the lattice
(30.7)
= {j1 + k2 : j, k Z}.
(z; ) =
X
1
1
1
+
.
z2
(z )2
2
06=
1
z
1
2 C 3,
2
(z )

for  2z.
To verify that
(30.10)
(z + ; ) = (z; ),
0 (z; ) = 2
1
,
(z )3
0 (z + ; ) = 0 (z; ),
Hence
(30.13)
(z + ; ) (z; ) = c(),
Now (30.8) implies (z; ) = (z; ). Hence, taking z = /2 in (30.13) gives c() = 0
for all , and we have (30.10).
200
Another analogy with (30.4) leads us to look at the function (not to be confused with
the Riemann zeta function)
(30.14)
(z; ) =
X 1
1
1
z
+
+ + 2 .
z
z
06=
We note that the sum here is obtained from the sum in (30.8) (up to sign) by integrating
from 0 to z along any path that avoids the poles. This is enough to establish convergence
of (30.14) in C \ , and we have
0 (z; ) = (z; ).
(30.15)
In view of (30.10), we hence have
(30.16)
(z + ; ) (z; ) = (),
a,b (z; ) = (z a; ) (z b; ),
(z + ) = (z),
Furthermore, this function has simple poles at (ip + ) (ip + 2 + ), and the residues at
ip and at ip + 2 cancel. Thus there exist constants A and B such that
(30.19)
The constants A and B can be evaluated by taking z = 0, 1, though the resulting formulas
give A and B in terms of special values of (z; ) rather than in elementary terms.
Exercises
1. Setting z = 1/2 in (30.2), show that
X
1
2
=
.
2
n
6
n=1
201
P
n1
nk for
(z n)(`+2) .
n=
(2k2)
1
2
2k
P
n=1
n2k .
n1,odd
n=1
= (2k 1)!22k+1
2k
+2
2k
n2k
n=1
n1,odd
1A. Set F (z) = ( cot z) 1/z, and use (30.4) to compute F (`) (0). Show that, for z < 1,
cot z =
X
1
2
(2k)z 2k1 ,
z
(2k) =
k=1
n2k .
n=1
1 ez + 1
1 X
Bk 2k1
=
+
(1)k1
z
,
z
2 e 1
z
(2k)!
k=1
with Bk (called the Bernoulli numbers) rational numbers for each k. Note that
e2iz + 1
1
= cot z.
2iz
e
1
i
Deduce from this and Exercise 1A that, for k 1,
2(2k) = (2)2k
Bk
.
(2k)!
202
Using
G(z) =
1 X
+
an z 2n1 ,
z n=1
compute the Laurent series expansions of G0 (z) and G(z)2 and deduce that a1 = 2 /3,
while, for n 2,
n1
1 X
an =
an` a` .
2n + 1
`=1
In concert with Exercise 1A, show that (2) = 2 /6, (4) = 4 /90, and also compute (6)
and (8).
2. Set
F (z) = z
Y
n=1
Show that
z2
1 2 .
n
F 0 (z)
1 X 2z
= +
.
F (z)
z n=1 z 2 n2
f (z + ) = f (z),
yields a meromorphic function on the torus T , defined by (26.14). Show that if such f
has no poles then it must be constant.
We say a parallelogram P C is a period parallelogram for a lattice (of the form (30.7))
provided it has vertices of the form p, p + 1 , p + 2 , p + 1 + 2 . Given a meromorphic
function f satisfying (30.20), pick a period parallelogram P whose boundary is disjoint
from the set of poles of f .
4. Show that
Z
f (z) dz = 0.
P
203
Deduce that
Respj (f ) = 0.
pj P
Deduce that if f has just one pole in P then that pole cannot be simple.
5. For defined by (30.14), show that, if Im(2 /1 ) > 0,
Z
(30.21)
(z; ) dz = (1 )2 (2 )1 = 2i.
P
( + 0 ) = () + ( 0 ),
, 0 .
Show that if , /2
/ , then
() = 2(/2; ).
7. Apply Greens theorem
ZZ
Z
g
f
dx dy = (f dx + g dy)
x y
(1 ) 2 (2 ) 1 = 2iI(),
with
ZZ
(30.24)
I() = lim
0
P\D (0)
(z; ) dx dy,
assuming P is centered at 0.
8. Solve the pair of equations (30.21) and (30.23) for (1 ) and (2 ). Use this in
concert with (30.22) to show that
(30.25)
() =
1
I() + ,
A(P)
204
nj pj and define
nj .
11. Show that  = 2 dim M (T ) = 2, and that this space is spanned by 1 and p1 ,p2
if n1 = n2 = 1, and by 1 and (z p1 ) if n1 = 2.
Hint. Use Exercise 4.
12. Show that
(30.27)
 = k 2 = dim M (T ) = k.
Hint. Argue by induction on k, noting that you can augment  by 1 either by adding
another pj or by increasing some positive nj by 1.
205
(31.1)
To see this, note that both sides have simple poles at z = . (As will be shown below, the
zeros of 0 (z) satisfy 2 .) The factor 0 () makes the poles cancel, so the difference
is entire, hence constant. Both sides vanish at z = 0, so this constant is zero. We also note
that
(31.2)
a,b (z) = , (z ),
a+b
ab
, =
.
2
2
As long as a b
/ , (31.1) applies, giving
(31.3)
a,b (z) =
0 ()
2(),
(z ) ()
a+b
ab
, =
.
2
2
We now prove the result on the zeros of 0 (z) stated above. Assume has the form
(30.7).
Proposition 31.1. The three points 1 /2, 2 /2 and (1 +2 )/2 are (mod ) all the zeros
of 0 (z).
Proof. Symmetry considerations (oddness of 0 (z)) imply 0 (z) = 0 at each of these three
points. Since 0 (z) has a single pole of order 3 in a period parallelogram, these must be
all the zeros. (Cf. Exercise 1 below to justify this last point.)
The general result hinted at above is the following.
Proposition 31.2. Let f be an elliptic function with period lattice . There exist rational
functions Q and R such that
(31.4)
Proof. First assume f is even, i.e., f (z) = f (z). The product of f (z) with factors of
the form (z) (a) lowers the degree of a pole of f at any point a
/ , so there exists
a polynomial P such that g(z) = P ((z))f (z) has poles only in . Note that g(z) is
206
also even. Then there exists a polynomial P2 such that g(z) P2 ((z)) has its poles
annihilated. This function must hence be constant. Hence any even elliptic f must be a
rational function of (z).
On the other hand, if f (z) is odd, then f (z)/0 (z) is even, and the previous argument
applies, so a general elliptic function must have the form (31.4).
The right side of (31.3) does not have the form (31.4), but we can come closer to this
form via the identity
(31.5)
(z ) = (z) () +
1 0 (z) + 0 () 2
.
4 (z) ()
This identity can be verified by showing that the difference of the two sides is pole free and
vanishes at z = 0. The right side of (31.5) has the form (31.4) except for the occurrence
of 0 (z)2 , which we will dispose of shortly.
Note that 0 (z)2 is even, with poles (of order 6) on . We can explicitly write this as
P ((z)), as follows. Set
(31.6)
ej =
j
,
2
j = 1, 2, 3,
In fact, both sides of (31.7) have poles of order 6, precisely at points of . Furthermore,
by Proposition 31.1, the zeros of 0 (z)2 occur precisely at z = j (mod ), each zero
having multiplicity 2. We also see that the right side of (31.7) has a double zero at
z = j , j = 1, 2, 3. So the quotient is entire, hence constant. The factor 4 arises by
examining the behavior as z 0.
The identity (31.7) is a differential equation for (z). Separation of variables yields
(31.8)
1
2
( e1 )( e2 )( e3 )
= z + c.
207
We can get more insight into the differential equation (31.7) by comparing Laurent
series expansions of the two sides about z = 0. First, we can deduce from (30.8) that
(31.9)
(z) =
1
+ az 2 + bz 4 + .
2
z
Of course, only even powers of z arise. Regarding the absence of the constant term and
the values of a and b, see Exercise 3 below. We have
(31.10)
a=3
X
06=
1
,
4
b=5
06=
1
.
6
Hence
(31.11)
0 (z) =
2
+ 2az + 4bz 3 + .
z3
1 0 2
1
2a
(z) = 6 2 4b + ,
4
z
z
while
((z) e1 )((z) e2 )((z) e3 )
(31.13)
where
1 = e1 + e2 + e3 ,
(31.14)
2 = e1 e2 + e2 e3 + e3 e1 ,
3 = e1 e2 e3 .
e1 + e2 + e3 = 0.
It also gives
(31.16)
e1 e2 + e2 e3 + e1 e3 = 5a,
e1 e2 e3 = 7b,
where a and b are given by (31.10). Hence we can rewrite the differential equation (31.7)
as
(31.17)
208
where g2 and g3 , known as the Weierstrass invariants of the lattice , are given by
(31.18)
g2 = 60
06=
1
,
4
g3 = 140
X
06=
1
.
6
Exercises
1. Assume f is meromorphic (and not identically zero) on T = C/. Show that the
number of poles of f is equal to the number of zeros of f , counting multiplicity.
Hint. Let bound a period parallelogram, avoiding the zeros and poles of f , and examine
1
2i
f 0 (z)
dz.
f (z)
+ z; =
z; ,
2
2
z.
X
1
1
1
(z) = (z) 2 =
2 ,
z
(z )2
\0
X
1 (k)
(0) = (k + 1)
(k+2) .
k!
\0
1
00 (z) = 6(z)2 g2 .
2
209
(z) =
X
1
+
bn z 2n ,
2
z
n1
where bn are given by (31.21), with k = 2n. Deduce from Exercise 5 that for n 3,
(31.24)
bn =
n2
X
3
bk bnk1 .
(2n + 3)(n 2)
k=1
In particular, we have
b3 =
and
b5 =
1 2
b ,
3 1
b4 =
3
b1 b2 ,
11
1 2
1 2 2 3
(b2 + 2b1 b3 ) =
b + b .
13
13 2 3 1
n =
X
\0
1
,
2n
then for n 3,
(31.26)
n = Pn (2 , 3 ),
2 =
1
(e1 e2 + e2 e3 + e1 e3 ),
15
3 =
1
e1 e2 e3 .
35
Note that bn = (2n + 1)n+1 . Note also that gk in (31.17)(31.18) satisfy g2 = 602 and
g3 = 1403 .
8. Given f as in Exercise 1, show that
(31.28)
1
2i
f 0 (z)
dz Z,
f (z)
210
On the one hand, Cauchys integral theorem (compare (5.19)) implies (31.30) is equal
to the left side of (31.29), provided pj and qj are all in the period domain. On the other
hand, if consists of four consecutive line segments, 1 , 2 , 3 , 4 , periodicity of f 0 (z)/f (z)
implies that (31.30) equals
Z 0
Z 0
2
1
f (z)
f (z)
(31.31)
dz +
dz.
2i
f (z)
2i
f (z)
1
Use Exercise 8 to deduce that the coefficients of 1 and 2 in (31.31) are integers.
10. Deduce from (31.5) that
(u) 0 (u) 1
u + v + w = 0 det (v) 0 (v) 1 = 0.
(w) 0 (w) 1
(31.32)
(2z) =
1 00 (z) 2
2(z).
4 0 (z)
13. Use (31.3) and (31.5), plus (31.7), to write a,b (z) (as in (31.3)) in the form (31.4), i.e.,
a,b (z) = Q((z)) + R((z))0 (z).
211
(x, t) =
en t e2inx ,
nZ
1 2
=
.
t
4 x2
(32.2)
(32.3)
This identity will ultimately lead us to a connection with (z). In addition, we have
2
1 X
(1)n en t e2inx ,
x + ,t =
2
(32.4)
nZ
and
(32.5)
X
2
i
x + t, t = et/4
e(n+1/2) t e2inx ,
2
nZ
which introduces series related to but slightly different from that in (32.1).
Following standard terminology, we set t = i , with Im > 0, and denote (z, i )
by
(32.6)
3 (z, ) =
en
i 2niz
nZ
p2n q n ,
nZ
where
p = eiz ,
(32.7)
q = ei .
4 (z, ) =
(1)n en
nZ
i 2inz
(1)n p2n q n ,
nZ
212
and
(32.9)
1 (z, ) = i
(1)n e(n1/2)
i (2n1)iz
=i
nZ
and finally
(32.10)
2 (z, ) =
nZ
e(n1/2)
i (2n1)iz
nZ
p2n1 q (n1/2) .
nZ
We will explore these functions, with the goal of relating them to (z). These results are
due to Jacobi; we follow the exposition of [MM].
To begin, we record how j (z + ) is related to k (z) for various values of . Here and
(usually) below we will suppress the and denote j (z, ) by j (z), for short. In the table
below we use
a = p1 q 1/4 = eizi /4 ,
b = p2 q 1 = e2izi .
1
2
3
4
z + 1/2
z + /2
z + 1/2 + /2
z+1
z+
z+1+
2
1
4
3
ia4
a3
a2
ia1
a3
ia4
ia1
a2
1
2
3
4
b1
b2
b3
b4
b1
b2
b3
b4
where
(32.13)
= {k + ` : k, ` Z}.
Gj (z) =
0j (z)
j (z)
satisfies
(32.15)
Gj (z + 1) = Gj (z),
Gj (z + ) = Gj (z) 2i.
To relate the functions Fjk to previously studied elliptic functions, we need to know the
zeros of k (z). Here is the statement:
213
(32.16)
1
2 (z) = 0 z + ,
2
1
3 (z) = 0 z + + , 4 (z) = 0 z + .
2 2
2
1 (z) = 0 z ,
Proof. In view of the relations tabulated above, it suffices to treat 1 (z). We first note
that
(32.17)
1 (z) = 1 (z).
2 j 4 = j (z) = j (z).
(32.20)
0 (0) (z) 2
2
1
(z) = e1 +
1 (z) 2 (0)
0 (0) (z) 2
3
1
= e2 +
1 (z) 3 (0)
0 (0) (z) 2
4
1
= e3 +
.
1 (z) 4 (0)
Proof. We have from (32.11)(32.13) that each function Pj (z) on the right side of (32.20)
is periodic. Also Proposition 32.1 implies each Pj has poles of order 2, precisely on
214
. Furthermore, we have arranged that each such function has leading singularity z 2
as z 0, and each Pj is even, by (32.17) and (32.18), so the difference (z) Pj (z) is
constant for each j. Evaluating at z = 1/2, (1 + )/2, and /2, respectively, shows that
these constants are zero, and completes the proof.
Part of the interest in (32.20) is that the series (32.6)(32.10) for the theta functions are
extremely rapidly convergent. To complete this result, we want to express the quantities
ej in terms of theta functions. The following is a key step.
Proposition 32.3. In the setting of Proposition 32.2,
0 (0) (0) 2
4
1
= 2 4 (0)4 ,
2 (0)3 (0)
0 (0) (0) 2
3
1
e1 e3 =
= 2 3 (0)4 ,
2 (0)4 (0)
0 (0) (0) 2
2
1
e2 e3 =
= 2 2 (0)4 .
3 (0)4 (0)
e1 e2 =
(32.21)
Proof. To get the first part of the first line, evaluate the second identity in (32.20) at
z = 1/2, to obtain
0 (0) (1/2) 2
3
1
,
e1 e2 =
1 (1/2) 3 (0)
and then consult the table to rewrite 3 (1/2)/1 (1/2). Similar arguments give the first
identity in the second and third lines of (32.21). The proof of the rest of the identities
then follows from the next result.
Proposition 32.4. We have
(32.22)
1 (2z) = 2
215
(32.25)
Now, from (32.2) we have
j
1 2 j
=
,
4i z 2
(32.26)
and computing
(32.27)
i
h
log 2 (0, ) + log 3 (0, ) + log 4 (0, ) log 01 (0, )
Thus
01 (0) = A2 (0)3 (0)4 (0),
(32.29)
01 (0) 2q 1/4 ,
2 (0) 2q 1/4 ,
3 (0) 1,
4 (0) 1,
e1 + e2 + e3 = 0
2
3 (0)4 + 4 (0)4 ,
3
2
e2 =
2 (0)4 4 (0)4 ,
3
2
e3 =
2 (0)4 + 3 (0)4 .
3
e1 =
(32.33)
216
Thus we have an efficient method to compute (z; ) when has the form (32.13). To
pass to the general case, we can use the identity
1 z
(32.34)
(z; a) = 2 ; .
a
a
See Appendix K for more on the rapid evaluation of (z; ).
Exercises
1. Show that
d 01 (z)
= a(z) + b,
dz 1 (z)
(32.35)
a = 1,
b = e1 +
where 1 = 1, 2 = .
2. In the setting of Exercise 1, deduce that a,b (z; ), given by (30.17), satisfies
01 (z a) 01 (z b)
1 (z a) 1 (z b)
d
1 (z a)
=
log
.
dz
1 (z b)
a,b (z; ) =
(32.37)
3. Show that, if a 6= ej ,
(32.38)
1
= A, (z) + B,
(z) a
A=
1
0 ()
Identify B.
4. Give a similar treatment of 1/((z) a) for a = ej . Relate these functions to (z
j ),
with
j found from (32.19).
5. Express g2 and g3 , given in (31.17)(31.18), in terms of theta functions.
Hint. Use Exercise 7 of 31, plus Proposition 32.5
217
q() = ( e1 )( e2 )( e3 ).
e1 + e2 + e3 = 0,
(33.3)
R(,
q()) d,
where R(, ) is a rational function of its arguments. The relevance of (31.8) is reinforced
by the following result.
Proposition 33.1. Given distinct ej satisfying (33.2), there exists a lattice , generated
by 1 , 2 C, linearly independent over R, such that if (z) = (z; ), then
(33.4)
= ej ,
1 j 3,
where 3 = 1 + 2 .
Given this result, we have from (31.7) that
0 (z)2 = 4q((z)),
(33.5)
and hence, as in (31.8),
(33.6)
1
2
(z)
(z0 )
d
p
= z z0 ,
q()
mod .
The problem of proving Proposition 33.1 is known as the Abel inversion problem. The
proof requires new tools, which will be provided in 34. We point out here that there is
no difficulty in identifying what the lattice must be. We have
(33.7)
1
2
ej
d
j
p
,
=
2
q()
mod ,
218
by (33.6). One can also verify directly from (33.7) that if the branches are chosen appropriately then 3 = 1 + 2 . It is not so clear that if is constructed directly from (33.7)
then the values of (z; ) at z = j /2 are given by (33.4), unless one already knows that
Proposition 33.1 is true.
Given Proposition 33.1, we can rewrite the elliptic integral (33.3) as follows. The result
depends on the particular path 01 from 0 to 1 and on the particular choice of path
01 in C/ such that maps 01 homeomorphically onto 01 . With these choices, (33.3)
becomes
Z
1
(33.8)
R (z), 0 (z) 0 (z) dz,
2
01
z1
(33.9)
z0
1
R (z), 0 (z) 0 (z) dz,
2
where z0 and z1 are the endpoints of 01 , satisfying (zj ) = j . It follows from Proposition
31.2 that
1 0 0
(33.10)
R (z), (z) (z) = R1 ((z)) + R2 ((z))0 (z),
2
for some rational functions Rj (). In fact, one can describe computational rules for producing such Rj , by using (33.5). Write R(, ) as a quotient of polynomials in (, ) and
use (33.5) to obtain that the left side of (33.10) is equal to
P1 ((z)) + P2 ((z))0 (z)
,
Q1 ((z)) + Q2 ((z))0 (z)
(33.11)
for some polynomials Pj (), Qj (). Then multiply the numerator and denominator of
(33.11) by Q1 ((z)) Q2 ((z))0 (z) and use (33.5) again on the new denominator to
obtain the right side of (33.10).
The integral of (33.3) is now transformed to the integral of the right side of (33.10).
Note that
Z z1
Z 1
0
(33.12)
R2 ((z)) (z) dz =
R2 () d, j = (zj ).
z0
219
by (30.15), where (z) = (z; ) is given by (30.14). See (32.35)(32.36) for a formula in
terms of theta functions. Next, differentiating (33.5) gives (as mentioned in Exercise 5 of
31)
1
00 (z) = 2q 0 ((z)) = 6(z)2 g2 ,
2
(33.15)
so
Z
(33.16)
z1
z0
g2
(z1 z0 ).
2
We can integrate (z)k+2 for k N via the following inductive procedure. We have
(33.17)
d 0
(z)(z)k = 00 (z)(z)k + k0 (z)2 (z)k1 .
dz
Apply (33.15) to 00 (z) and (33.5) (or equivalently (31.17)) to 0 (z)2 to obtain
d 0
(z)(z)k ) = (6 + 4k)(z)k+2 (3 + k)g2 (z)k kg3 (z)k1 .
dz
Rz
From here the inductive evaluation of z01 (z)k+2 dz, for k = 1, 2, 3, . . . , is straightforward.
To analyze (33.13) for a general rational function R1 (), we see upon making a partial
fraction decomposition that it remains to analyze
Z z1
(33.19)
((z) a)` dz,
(33.18)
z0
for ` = 1, 2, 3, . . . . One can also obtain inductive formulas here, by replacing (z)k by
((z) a)k in (33.18) and realizing that k need not be positive. We get
(33.20)
d 0
(z)((z) a)k = 00 (z)((z) a)k + k0 (z)2 ((z) a)k1 .
dz
Now write
(33.21)
where
(33.22)
j =
q (j) (a)
,
j!
Aj =
q (j+1) (a)
,
j!
220
to obtain
(33.23)
d 0
(z)((z) a)k
dz
= (2A2 + 4k3 )((z) a)k+2 + (2A1 + 4k2 )((z) a)k+1
+ (2A0 + 4k1 )((z) a)k + 4k0 ((z) a)k1 .
Note that
(33.24)
0 = q(a),
Thus, if a is not equal to ej for any j and if we know the integral (33.19) for integral
` k (for some negative integer k), we can compute the integral for ` = 1 k, as long
as k 6= 0. If a = ej for some j, and if we know (33.19) for integral ` k 1, we can
compute it for ` = k, since then q 0 (a) 6= 0.
At this point, the remaining case of (33.19) to consider is the case ` = 1, i.e.,
Z
z1
(33.25)
z0
dz
.
(z) a
(33.26)
0
d
p
,
q()
1
0
d
( a) p
,
q()
1
d
p
.
a q()
These are called, respectively, elliptic integrals of the first, second, and third kind. The
material given above expresses the general elliptic integral (33.3) in terms of these cases.
There is another family of elliptic integrals, namely those of the form
Z
p
(33.27)
I = R(, Q()) d,
where R(, ) is a rational function of its arguments and Q() is a fourth degree polynomial:
(33.28)
Q() = ( a0 )( a1 )( a2 )( a3 ),
with aj C distinct. Such integrals can be transformed to integrals of the form (33.3),
via the change of variable
(33.29)
1
,
a0
d =
1
d.
2
221
One has
(33.30)
1
11
1
1
Q
+ a0 =
+ a0 a1
+ a0 a2
+ a0 a3
A
= 4 ( e1 )( e2 )( e3 ),
where
(33.31)
ej =
1
.
aj a0
Then we have
Z
(33.32)
I=
+ a0 ,
1
A p
q(
)
d,
2
2
with q( ) as in (33.1). After a further coordinate translation, one can alter ej to arrange
(33.2).
Elliptic integrals are frequently encountered in many areas of mathematics. Here we
give two examples, one from differential equations and one from geometry.
Our first example involves the differential equation for the motion of a simple pendulum,
which takes the form
(33.33)
d2
+ g sin = 0,
dt2
where ` is the length of the pendulum g the acceleration of gravity (32 ft./sec.2 on the
surface of the earth), and is the angle the pendulum makes with the downwardpointing
vertical axis. The total energy of the pendulum is proportional to
1 d 2 g
E=
cos .
2 dt
`
(33.34)
Applying d/dt to (33.34) and comparing with (33.33) shows that E is constant for each
solution to (33.33), so one has
r
1 d
g
(33.35)
= E + cos ,
`
2 dt
or
Z
(33.36)
d
= 2t + c,
E + a cos
222
d
= 2 t + c,
(33.39)
p
(E + a)(1 2 )
which is an integral of the form (33.3).
In our next example we produce a formula for the arc length L() of the portion of the
ellipse
(33.40)
(33.41)
= b2 + c2 sin2 t,
with c2 = a2 b2 , so
Z
(33.42)
L() =
p
b2 + c2 sin2 t dt.
Exercises
1. Using (33.7) and the comments that follow it, show that, for j = 1, 2,
(33.44)
1
2
e3
ej
d
j 0
p
,
=
2
q()
mod ,
223
where j 0 = 2 if j = 1, j 0 = 1 if j = 2.
2. Setting ekj = ek ej , show that
(33.45)
1
2
e1 +
e1
X
d
1
1/2 1/2
1
k+`+1/2
p
=
2 e12 e13
k
`
ek12 e`13 k + ` + 1/2
q()
k,`=0
is a convergent power series provided  < min(e1 e2 , e1 e3 ). Using this and variants
to integrate from ej to ej + for j = 2 and 3, find convergent power series for j /2 (mod
).
3. Given k 6= 1, show that
Z
Z
d
d
1
p
p
(33.46)
= p
,
(1 2 )(k 2 2 )
2(1 k 2 )
q( )
with
=
1
1
1
q( ) =
.
2
1k
1+k
1
,
+1
/2
3 = 2
= 2I e3 e2 , e3 e1 ,
0
where
Z
(33.49)
/2
I(r, s) =
0
d
2
r2 sin + s2 cos2
I(r, s) =
,
2M (s, r)
224
(ak+1 , bk+1 ) =
a + b p
k
k
, ak bk .
2
Show that
a0 a1 a2 b2 b1 b0 .
Show that
the latter identity being the definition of M (a, b). Show also that
a2k+1 b2k+1 =
1
(ak bk )2 ,
4
ak+1 bk+1 =
(ak bk )2
.
8ak+2
hence
(33.53)
I(r, s) = I
rs,
r + s
.
2
Hint. Show that (33.54) I(r, s) = I(m, m), with m = M (s, r), and evaluate I(m, m).
7. Take the following steps to prove (33.54). Show that you can make the change of
variable from to , with
(33.55)
sin =
2s sin
,
(s + r) + (s r) sin2
,
2
and obtain
Z
(33.56)
/2
I(r, s) =
0
2 d
q
.
4rs sin2 + (s + r)2 cos2
225
3 =
.
M ( e3 e1 , e3 e2 )
e3
1 =
( e1 )( e2 )( e3 )
e2
/2
(33.58)
= 2i
.
M ( e3 e1 , e2 e1 )
d
2
1 2 sin
dx
=
0
(1
x2 )(1
2 x2 )
dx
p
p
=
,
(3.59)
(1 x2 )(1 2 x2 )
2M (1, 1 2 )
0
if (1, 1).
11. Parallel to Exercise 10, show that
Z /2
Z 1
d
dx
p
p
,
=
(1 x2 )(1 + 2 x2 )
0
0
1 + 2 sin2
and deduce that
Z
(3.60)
0
dx
p
p
=
,
(1 x2 )(1 + 2 x2 )
2M ( 1 + 2 , 1)
if R. A special case is
Z
(3.61)
0
dx
.
=
2M ( 2, 1)
1 x4
226
q()
q() = ( e1 )( e2 )( e3 ),
: M S 2 ,
and discuss some important properties of M and . We will then use this construction
to prove Proposition 33.1. Material developed below will use some basic results on manifolds, particularly on surfaces, which are generally covered in beginning topology courses.
Background may be found in [Mun] and [Sp], among other places. See also Appendix C
for some background.
To begin, we set e4 = in the Riemann sphere C {}, identified with S 2 in 26.
Reordering if necessary, we arrange that the geodesic 12 from e1 to e2 is disjoint from the
geodesic 34 from e3 to e4 . We slit S 2 along 12 and along 34 , obtaining X, a manifold
with boundary, as illustrated in the top right portion of Fig. 34.1. Now
(34.3)
M = X1 X2 / ,
where X1 and X2 are two copies of X, and the equivalence relation identifies the upper
boundary of X1 along the slit 12 with the lower boundary of X2 along this slit and viceversa, and similarly for 34 . This is illustrated in the middle and bottom parts of Fig. 34.1.
The manifold M is seen to be topologically equivalent to a torus.
The map : M S 2 in (34.2) is tautological. It is twotoone except for the four
points pj = 1 (ej ). Recall the definition of a Riemann surface given in 26, in terms
of coordinate covers. The space M has a unique Riemann surface structure for which
is holomorphic. A coordinate taking a neighborhood of pj in M bijectively onto a
neighborhood of the origin in C is given by j (x) = ((x) ej )1/2 , for 1 j 3, with
(x) S 2 C {}, and a coordinate mapping a neighborhood of p4 in M bijectively
1/2
onto a neighborhood of the origin in C is given
.
pby 4 (x) =2 (x)
Now consider the doublevalued form d/ q() on S , having singularities at {ej }.
This pulls back to a singlevalued 1form on M . Noting that if w2 = then
(34.4)
d
= 2 dw,
d
p = 2 dw,
3
we see that is a smooth, holomorphic 1form on M , with no singularities, and also that
has no zeros on M . Using this, we can prove the following.
227
: M C/0 .
p0
: M C/1 .
e1 + e2 + e3 = 0.
= ej , 1 j 3,
2
where 3 = 1 + 2 .
Proof. We have from (34.2) and (34.6) a holomorphic map
(34.11)
: C/0 S 2 ,
(z + ) = (z),
0 .
228
a C , b C,
(z) = a0 (z) + b,
where 0 (z) = (z; 0 ). Hence 0 (z) = a00 (z), so by Proposition 31.1 we have
(34.14)
0 (z) = 0 z =
0j
,
2
mod 0 ,
ej = a0
0j
+ b.
Now if e0j = 0 (0j /2), we have by (31.15) that e01 + e02 + e03 = 0, so (34.9) yields
(34.16)
b = 0.
(z; ) = a (a1/2 z; 0 ).
q() = ( e1 ) ( em ),
where ej C are distinct, and m 2. If m = 2g + 1, one has slits from e2j1 to e2j , for
j = 1, . . . , g, and a slit from e2g+1 to , which we denote e2g+2 . If m = 2g + 2, one has
slits from e2j1 to e2j , for j = 1, . . . , g + 1. Then X is constructed by opening the slits,
and M is constructed as in (34.3). The picture looks like that in Fig. 34.1, but instead of
two sets of pipes getting attached, one has g + 1 sets. One gets a Riemann
p surface M with
g holes, called a surface of genus g. Again the doublevalued form d/ q() on S 2 pulls
back to a singlevalued 1form on M , with no singularities, except when m = 2 (see the
exercises). If m = 4 (so again g = 1), has no zeros. If m 5 (so g 2), has a zero
at 1 (). Proposition 34.1 extends to the case m = 4. If m 5 the situation changes.
It is a classical result that M is covered by the disk D rather than by C. The pullback
of to D is called an automorphic form. For much more on such matters, and on more
general constructions of Riemann surfaces, we recommend [FK] and [MM].
We end this section with a brief description of a Riemann surface, conformally equivalent
to M in (34.3), appearing as a submanifold of complex projective space CP2 . More details
on such a construction can be found in [Cl] and [MM].
229
To begin, we define complex projective space CPn as (Cn+1 \ 0)/ , where we say z and
z 0 Cn+1 \ 0 satisfy z z 0 provided z 0 = az for some a C . Then CPn has the structure
of a complex manifold. Denote by [z] the equivalence class in CPn of z Cn+1 \ 0. We
note that the map
(34.19)
: CP1 C {}
given by
(34.20)
z2 6= 0,
([(1, 0)]) = ,
One can show that Me is a smooth complex submanifold of CP2 , possessing then the
structure of a compact Riemann surface. An analogue of the map (34.2) is given as
follows.
Set p = [(1, 0, 0)] CP2 . Then there is a holomorphic map
(34.22)
: CP2 \ p CP1 ,
given by
(34.23)
e : Me CP1 C {} S 2 ,
given by (34.22) on Me \ p and taking p to [(1, 0)] CP1 , hence to C {}. This map
can be seen to be a 2to1 branched covering, branching over B = {e1 , e2 , e3 , }. Given
q C, q
/ B, and a choice r 1 (q) M and re 1
e (q) Me , there is a unique
holomorphic diffeomorphism
(34.25)
such that (r) = re and = e .
: M Me ,
230
Exercises
1. Show that the covering map in (34.8) is actually a diffeomorphism, and hence 0 = 1 .
2. Suppose 0 and 1 are two lattices in C such that T0 and T1 are conformally
equivalent, via a holomorphic diffeomorphism
(34.26)
f : C/0 C/1 .
Show that f lifts to a holomorphic diffeomorphism F of C onto itself, such that F (0) = 0,
and hence that F (z) = az for some a C . Deduce that 1 = a0 .
3. Consider the upper halfplane U = { C : Im > 0}. Given U , define
(34.27)
( ) = {m + n : m, n Z}.
(34.28)
( ) = + 1.
(( )) = 1 ( ),
(( )) = ( ).
C/( ) C/( 0 ),
0 = ( ),
for some G.
6. Show that the group G consists of linear fractional transformations of the form
a + b
a b
(34.32)
LA ( ) =
, A=
,
c d
c + d
where a, b, c, d Z and det A = 1, i.e., A Sl(2, Z). Show that
G Sl(2, Z)/{I}.
231
In Exercises 78, we make use of the covering map : U C \ {0, 1}, given by (25.5),
and results of Exercises 18 of 25, including (25.10)(25.11), i.e.,
(34.33)
(( )) =
1
,
( )
(( )) = 1 ( ).
, 0 U,
w = ( ), w0 = ( 0 ) C \ {0, 1},
we have
0 w0 = F (w) for some F G,
(34.35)
H(w) =
4 (w2 w + 1)3
,
27 w2 (w 1)2
1
w
= H(w),
H(1 w) = H(w),
show that
(34.38)
0 H ( 0 ) = H ( ).
q() when
q() = ( e1 )( e2 ),
232
p
10. Show that the doublevalued form d/ q() on S 2 pulls back to a singlevalued form
on M . Using (34.4), show that is a smooth nonvanishing form except at {p1 , p2 } =
1 (). In a local coordinate system about pj of the form j (x) = (x)1 , use a variant
of (34.4)(34.5) to show that has the form
= (1)j
(34.41)
g(z)
dz,
z
: M \ {p1 , p2 } C/L.
(34.45)
(34.46)
14. Note that with a linear change of variable we can arrange ej = (1)j in (34.40). Relate
the results of Exercises 913 to the identity
Z z
(1 2 )1/2 d = sin1 z (mod 2Z).
(34.47)
0
233
The third of these properties is called the triangle inequality. An example of a metric space
is the set of rational numbers Q, with d(x, y) = x y. Another example is X = Rn , with
d(x, y) =
p
(x1 y1 )2 + + (xn yn )2 .
xj K, xj p X = p K.
234
We will establish various properties of compact metric spaces, and provide various equivalent characterizations. For example, it is easily seen that (A) is equivalent to:
(B)
235
Proof. Each U is a union of open balls, so it suffices to show that (A) implies the following:
(D)
236
Xj B2j (yj ),
each containing infinitely many points of S. One sees that (yj ) forms a Cauchy sequence.
If X is complete, it has a limit, yj z, and z is seen to be an accumulation point of S.
If Xj , 1 j m, is a finite collection of metric spaces, with metrics dj , we can define
a Cartesian product metric space
(A.4)
X=
m
Y
Xj ,
j=1
p
Another choice of metric is (x, y) = d1 (x1 , y1 )2 + + dm (xm , ym )2 . The metrics d and
are equivalent, i.e., there exist constants C0 , C1 (0, ) such that
(A.5)
x, y X.
237
Qm
j=1
Xj .
U open in Y = f 1 (U ) open in X.
238
sup S = max x,
inf S = min x,
sS
xS
when S is bounded. More generally, if S R is nonempty and bounded from above, say
S (, B], we can pick A < B such that S [A, B] is nonempty, and set
(A.8)
Similarly, if S R is nonempty and bounded from below, say S [A, ), we can pick
B > A such that S [A, B] is nonempty, and set
(A.9)
xX
If f is not bounded from above, we set sup f = +, and if f is not bounded from below,
we set inf f = .
Given a set X, f : X R, and xn x, we set
(A.11A)
lim sup f (xn ) = lim sup f (xk ) ,
n
n kn
and
(A.11B)
inf f (xk ) .
n kn
Not all continuous functions are uniformly continuous. For example, if X = (0, 1) R,
then f (x) = sin 1/x is continuous, but not uniformly continuous, on X. The following
result is useful, for example, in the development of the Riemann integral.
239
f (x ) f (y ) .
(A.15)
D(f, g) = sup d f (x), g(x) .
xX
This sup exists provided f (X) and g(X) are bounded subsets of Y, where to say B Y is
bounded is to say d : B B [0, ) has bounded image. In particular, this supremum
exists if X is compact. The following is a natural completeness result.
Proposition A.16. If X is a compact metric space and Y is a complete metric space,
then C(X, Y ), with the metric (A.9), is complete.
Proof. That D(f, g) satisfies the conditions to define a metric on C(X, Y ) is straightforward. We check completeness. Suppose (f ) is a Cauchy sequence in C(X, Y ), so, as
,
(A.16)
sup sup d f+k (x), f (x) 0.
k0 xX
(A.17)
sup d g(x), f (x) 0,
xX
i.e., f g uniformly. It remains only to show that g is continuous. For this, let xj x
in X and fix > 0. Pick N so that N < . Since fN is continuous, there exists J such
that j J d(fN (xj ), fN (x)) < . Hence
240
kaf k = a kf k,
kf + gk kf k + kgk,
given f, g V and a a scalar (in R or C). A vector space equipped with a norm is called a
normed vector space. It is then a metric space, with distance function D(f, g) = kf gk.
If the space is complete, one calls V a Banach space.
In particular, by Proposition A.16, C(X) is a Banach space, when X is a compact
metric space.
We next give a couple of slightly more sophisticated results on compactness. The following extension of Proposition A.11 is a special case of Tychonovs Theorem.
Q
Proposition A.17. If {Xj : j Z+ } are compact metric spaces, so is X = j=1 Xj .
Here, we can make X a metric space by setting
(A.18)
d(x, y) =
X
j=1
2j
where pj : X Xj is the projection onto the jth factor. It is easy to verify that, if x X,
then x y in X, as , if and only if, for each j, pj (x ) pj (y) in Xj .
Proof. Following the argument in Proposition A.11, if (x ) is an infinite sequence of points
in X, we obtain a nested family of subsequences
(A.19)
such that p` (xj ) converges in X` , for 1 ` j. The next step is a diagonal construction.
We set
(A.20)
= x X.
Then, for each j, after throwing away a finite number N (j) of elements, one obtains from
( ) a subsequence of the sequence (xj ) in (A.19), so p` ( ) converges in X` for all `.
Hence ( ) is a convergent subsequence of (x ).
The next result is a special case of Ascolis Theorem.
Proposition A.18. Let X and Y be compact metric spaces, and fix a modulus of continuity (). Then
(A.21)
C = f C(X, Y ) : d f (x), f (x0 ) d(x, x0 ) x, x0 X
is a compact subset of C(X, Y ).
Proof. Let (f ) be a sequence in C . Let be a countable dense subset of X, as in Corollary
A.5. For each x , (f (x)) is a sequence in Y, which hence has a convergent subsequence.
241
Using a diagonal construction similar to that in the proof of Proposition A.17, we obtain
a subsequence ( ) of (f ) with the property that (x) converges in Y, for each x ,
say
(A.22)
(x) (x),
(A.23)
+ d (x` ), (x)
/3 + /3 + /3.
Thus ( (x)) is Cauchy in Y for all x X, hence convergent. Call the limit (x), so we
now have (A.22) for all x X. Letting k in (A.23) we have uniform convergence of
to . Finally, passing to the limit in
(A.24)
gives C .
We want to restate Proposition A.18, bringing in the notion of equicontinuity. Given
metric spaces X and Y , and a set of maps F C(X, Y ), we say F is equicontinuous at a
point x0 X provided
(A.25)
Note that (A.26) is equivalent to the existence of a modulus of continuity such that
F C , given by (A.21). It is useful to record the following result.
Proposition A.19. Let X and Y be metric spaces, F C(X, Y ). Assume X is compact.
then
(A.27)
242
Proof. The argument is a variant of the proof of Proposition A.15. In more detail, suppose
there exist x , x0 X, > 0, and f F such that d(x , x0 ) 2 but
(A.28)
d(f (x ), f (x0 )) .
,
2
j N, g F,
contradicting (A.28).
Putting together Propositions A.18 and A.19 then gives the following.
Proposition A.20. Let X and Y be compact metric spaces. If F C(X, Y ) is equicontinuous on X, then it has compact closure in C(X, Y ).
We next define the notion of a connected space. A metric space X is said to be connected
provided that it cannot be written as the union of two disjoint nonempty open subsets.
The following is a basic class of examples.
Proposition A.21. Each interval I in R is connected.
Proof. Suppose A I is nonempty, with nonempty complement B I, and both sets are
open. Take a A, b B; we can assume a < b. Let = sup{x [a, b] : x A} This
exists, as a consequence of the basic fact that R is complete.
Now we obtain a contradiction, as follows. Since A is closed A. But then, since A
is open, there must be a neighborhood ( , + ) contained in A; this is not possible.
We say X is pathconnected if, given any p, q X, there is a continuous map : [0, 1]
X such that (0) = p and (1) = q. It is an easy consequence of Proposition A.21 that X
is connected whenever it is pathconnected.
The next result, known as the Intermediate Value Theorem, is frequently useful.
Proposition A.22. Let X be a connected metric space and f : X R continuous.
Assume p, q X, and f (p) = a < f (q) = b. Then, given any c (a, b), there exists z X
such that f (z) = c.
Proof. Under the hypotheses, A = {x X : f (x) < c} is open and contains p, while
B = {x X : f (x) > c} is open and contains q. If X is connected, then A B cannot be
all of X; so any point in its complement has the desired property.
The next result is known as the Contraction Mapping Principle, and it has many uses
in analysis. In particular, we will use it in the proof of the Inverse Function Theorem, in
Appendix B.
243
for some r < 1. (We say T is a contraction.) Then T has a unique fixed point x. For any
y0 X, T k y0 x as k .
Proof. Pick y0 X and let yk = T k y0 . Then d(yk , yk+1 ) rk d(y0 , y1 ), so
d(yk , yk+m ) d(yk , yk+1 ) + + d(yk+m1 , yk+m )
(A.31)
Exercises
1. If X is a metric space, with distance function d, show that
d(x, y) d(x0 , y 0 ) d(x, x0 ) + d(y, y 0 ),
and hence
d : X X [0, ) is continuous.
2. Let : [0, ) [0, ) be a C 2 function. Assume
(0) = 0,
0 > 0,
00 < 0.
Prove that if d(x, y) is symmetric and satisfies the triangle inequality, so does
(x, y) = (d(x, y)).
Hint. Show that such satisfies (s + t) (s) + (t), for s, t R+ .
3. Show that the function d(x, y) defined by (A.18) satisfies (A.1).
Hint. Consider (r) = r/(1 + r).
4. Let X be a compact metric space. Assume fj , f C(X) and
fj (x) % f (x),
x X.
244
m (x, y).
6. Let X and Y be compact metric spaces. Show that if F C(X, Y ) is compact, then F
is equicontinuous. (This is a converse to Proposition A.20.)
7. Recall that a Banach space is a complete normed linear space. Consider C 1 (I), where
I = [0, 1], with norm
kf kC 1 = sup f  + sup f 0 .
I
245
F (x + y) = F (x) + Ly + R(x, y)
with
(B.2)
kR(x, y)k
0 as y 0.
kyk
Fj
(B.3)
DF (x) =
,
xk
so that, if v = (v1 , . . . , vn )t , (regarded as a column vector) then
(B.4)
DF (x)v =
X F
k
xk
vk , . . . ,
X Fm
k
xk
vk
It will be shown below that F is differentiable whenever all the partial derivatives exist
and are continuous on O. In such a case we say F is a C 1 function on O. More generally,
F is said to be C k if all its partial derivatives of order k exist and are continuous. If F
is C k for all k, we say F is C .
In (B.2), we can use the Euclidean norm on Rn and Rm . This norm is defined by
(B.5)
1/2
kxk = x21 + + x2n
(B.6)
246
with
F (x + y) = F (x) +
DF (x + ty)y dt,
0
F (x + y) = F (x) +
F (x + zj ) F (x + zj1 ) = F (x) +
Aj (x, y)yj ,
(B.9)
j=1
j=1
Aj (x, y) =
0
F
x + zj1 + tyj ej dt,
xj
where z0 = 0, zj = (y1 , . . . , yj , 0, . . . , 0), and {ej } is the standard basis of Rn . Now (B.9)
implies F is differentiable on O, as we stated below (B.4). Thus we have established the
following.
Proposition B.1. If O is an open subset of Rn and F : O Rm is of class C 1 , then F
is differentiable at each point x O.
As is shown in many calculus texts, one can use the Mean Value Theorem instead of
the Fundamental Theorem of Calculus, and obtain a slightly sharper result.
For the study of higher order derivatives of a function, the following result is fundamental.
Proposition B.2. Assume F : O Rm is of class C 2 , with O open in Rn . Then, for
each x O, 1 j, k n,
(B.10)
F
F
(x) =
(x).
xj xk
xk xj
j,h f (x) =
1
f (x + hej ) f (x) ,
h
247
where {e1 , . . . , en } is the standard basis of Rn . The mean value theorem (for functions of
xj alone) implies that if j f = f /xj exists on O, then, for x O, h > 0 sufficiently
small,
(B.12)
for some j (0, 1), depending on x and h. Iterating this, if j (k f ) exists on O, then,
for x O and h > 0 sufficiently small,
k,h j,h f (x) = k (j,h f )(x + k hek )
(B.13)
k j,h f = j,h (k f ).
Clearly
(B.16)
j k f (x0 ) = k j f (x0 ).
248
(B.18)
f (u) = A F (p0 + u) q0 , A = DF (p0 )1 .
Then f (0) = 0 and Df (0) = I, the identity matrix. We show that f maps a neighborhood
of 0 onetoone and onto some neighborhood of 0. Proposition B.4 applies, so we know f
is onetoone on some neighborhood O of 0. We next show that the image of O under f
contains a neighborhood of 0.
Note that
(B.19)
f (u) = u + R(u),
R(0) = 0, DR(0) = 0.
f (u) = v.
Tv (u) = v R(u).
Tv (u) = u.
We look for a fixed point u = K(v) = f 1 (v). Also, we want to prove that DK(0) = I, i.e.,
that K(v) = v + r(v) with r(v) = o(kvk), i.e., r(v)/kvk 0 as v 0. If we succeed in
doing this, it follows easily that, for general x close to q0 , G(x) = F 1 (x) is defined, and
(B.23)
1
DG(x) = DF G(x)
.
249
for some r < 1 (we say T is a contraction), then T has a unique fixed point x.
In order to implement this, we consider
(B.25)
Tv : Xv Xv
with
(B.26)
Xv = {u : ku vk Av }
where we set
(B.27)
Av =
sup
kR(w)k.
kwk2kvk
We claim that (B.25) holds if kvk is sufficiently small. To prove this, note that Tv (u) v =
R(u), so we need to show that, provided kvk is small, u Xv implies kR(u)k Av . But
indeed, if u Xv , then kuk kvk + Av , which is 2kvk if kvk is small, so then
kR(u)k
sup
kR(w)k = Av .
kwk2kvk
kK(v) vk Av = o(kvk),
250
Tp M = Range D(x0 ),
: O Rnm Rn ,
(C.4)
1 (x, 0) = F (x), 0 , 1 (y, 0) = F 1 (y), 0 .
251
(C.5)
= F.
y = F (x).
In particular this implies that Range D(x0 ) = Range D(y0 ), so Tp M in (C.1) is independent of the choice of coordinate chart. It is called the tangent space to M at p.
We next define an object called the metric tensor on M . Given a coordinate chart
: O U , there is associated an m m matrix G(x) = gjk (x) of functions on O,
defined in terms of the inner product of vectors tangent to M :
n
X ` `
gjk (x) = D(x)ej D(x)ek =
=
,
xj xk
xj xk
(C.7)
`=1
(C.8)
We call (gjk ) the metric tensor of M, on U , with respect to the coordinate chart : O U .
Note that this matrix is positivedefinite. From a coordinateindependent point of view,
the metric tensor on M specifies inner products of vectors tangent to M , using the inner
product of Rn .
If we take another coordinate chart : U, we want to compare (gjk ) with H =
(hjk ), given by
(C.9)
(C.10)
or equivalently,
(C.11)
gjk (x) =
X Fi F`
hi` (y).
xj xk
i,`
252
where
(C.13)
(C.15)
so the right side of (C.14) is seen to be equal to (C.12), and our surface integral is well
defined, at least for f supported in a coordinate patch. More generally, if f : M R has
compact support, write it as a finite sum of terms, each supported on a coordinate patch,
and use (C.12) on each patch.
Let us pause to consider the special cases m = 1 and m = 2. For m = 1, we are
considering a curve in Rn , say : [a, b] Rn . Then G(x) is a 1 1 matrix, namely
G(x) = 0 (x)2 . If we denote the curve in Rn by , rather than M, the formula (C.12)
becomes
Z
Z b
(C.16)
f ds =
f (x) 0 (x) dx.
a
1 1
(C.18)
g = det
2 1
1 2
= 1 2 2 2 (1 2 )2 .
2 2
Recall that u v = u v  sin , where is the angle between u and v. Equivalently,
since u v = u v cos ,
(C.19)
u v2 = u2 v2 1 cos2 = u2 v2 (u v)2 .
253
gjk (x) = jk +
u u
.
xj xk
(C.21)
1/2
g = 1 + u2
.
1/2
(C.22)
Vn1 (M ) = dS =
1 + u(x)2
dx.
M
(C.23)
R : (0, ) S n1 Rn \ 0,
R(r, ) = r.
Let (h`m ) denote the metric tensor on S n1 (induced from its inclusion in Rn ) with respect
to some coordinate chart : O U S n1 . Then, with respect to the coordinate chart
: (0, ) O U Rn given by (r, y) = r(y), the Euclidean metric tensor can be
written
1
(C.25)
ejk =
.
r2 h`m
To see that the blank terms vanish, i.e., r xj = 0, note that (x) (x) = 1
xj (x) (x) = 0. Now (C.25) yields
(C.26)
e = rn1 h.
We therefore have the following result for integrating a function in spherical polar coordinates.
Z
Z hZ
i
(C.27)
f (x) dx =
f (r)rn1 dr dS().
Rn
S n1
254
ex dx = n/2 ,
(C.28)
Rn
which can be reduced to the case n = 2 and done there in polar coordinates. First note
that, whenever f (x) = (x), (C.27) yields
Z
Z
(x) dx = An1
(C.29)
(r)rn1 dr.
Rn
2
n/2
= An1
r 2 n1
1
dr = An1
2
es sn/21 ds,
where we used the substitution s = r2 to get the last identity. We hence have
(C.31)
An1 =
2 n/2
,
(n/2)
(z) =
es sz1 ds.
255
1 : U1 M,
(C.34)
G(x) = gjk (x) , gjk C k (U1 ),
which is taken to be symmetric and positive definite, generalizing the setup in (C.7). If
there is another covering of M by a coordinate chart 2 : U2 M , a positive definite
matrix H on U2 defines the same metric tensor on M provided
G(x) = DF (x)t H(y)DF (x),
(C.35)
for y = F (x),
(C.36)
We also say that G defines a metric tensor on U1 and H defines a metric tensor on U2 ,
and the diffeomorphism F : U1 U2 pulls H back to G.
Let : [a, b] U1 be a C 1 curve. The following is a natural generalization of (C.16),
defining the integral of a function f C(U1 ) over , with respect to arc length:
Z
Z b
h
i1/2
0
0
(C.37)
f ds =
f ((t)) (t) G((t)) (t)
dt.
a
f ds =
f(
(t)) 0 (t) H(
(t))
0 (t)
dt
a
h
i1/2
f ((t)) DF ((t)) 0 (t) H(
(t))DF ((t)) 0 (t)
dt
h
i1/2
f ((t)) 0 (t) G(x) 0 (t)
dt
=
(C.38)
=
Za
=
f ds,
256
the second identity by the chain rule 0 (t) = DF ((t)) 0 (t) and the third identity by (C.35).
Another property of this integral is parametrization invariance. Say : [, ] [a, b] is
an order preserving C 1 diffeomorphism and = : [, ] U1 . Then
Z
f ds =
h
i1/2
f ((t)) 0 (t) G((t)) 0 (t)
dt
(C.39)
Z
=
h
i1/2
dt
f ( (t)) 0 (t)2 0 ((t)) G( (t)) 0 ((t))
h
i1/2
0
0
f (( )) ( ) G(( )) ( )
d
Za
=
f ds,
the second identity by the chain rule 0 (t) = 0 (t) 0 ((t)) and the third identity via the
change of variable = (t), d = 0 (t) dt.
The arc length of these curves is defined by integrating 1. We have
Z bh
(C.40)
`G () =
i1/2
0 (t) G((t)) 0 (t)
dt,
`G () = `G () = `H (
).
In case
(C.44)
we have
(C.45)
257
or
(C.47)
ds = A(z) dz.
`G () =
Under the change of variable y = F (x), the formula (C.35) for the metric tensor H =
(hjk ) on U2 that pulls back to G under F : U1 U2 is equivalent to
(C.49)
X
j,k
X
j,k
dyj =
X Fj
`
x`
dx` .
258
D. Greens theorem
Here we prove Greens theorem, which was used in one approach to the Cauchy integral
theorem in 5.
Theorem D.1. If is a bounded region in R2 with piecewise smooth boundary, and
f, g C 1 (), then
ZZ
Z
g
f
dx dy = (f dx + g dy).
x y
(D.1)
ZZ
(D.2)
f
dx dy,
y
f dx =
and
Z
(D.3)
ZZ
g dy =
g
dx dy.
x
We will first prove (D.2) for domains of the form depicted in Fig. D.1 (which we will
call type I), then prove (D.3) for domains of the form depicted in Fig. D.2 (which we call
type II), and then discuss more general domains.
If is type I (cf. Fig. D.1), then
Z
(D.4)
f dx =
a
f (x, 0 (x)) dx
1 (x)
f
(x, y) dy,
y
1 (x)
0 (x)
f
(x, y) dy dx =
y
ZZ
f
dx dy,
y
259
Z
g dy =
g(1 (y)) dy
c
(D.7)
g(0 (y), y) dy
c
1 (y)
=
c
0 (y)
ZZ
g
(x, y), dx dy
x
g
dx dy,
x
ZZ
X Z Z g
g
dx dy =
dx dy.
y
y
j
j
R
On the other hand, if we sum the integrals j g dy, we get an integral over plus two
integrals over the interface between 1 and 2 . However, the latter two integrals cancel
out, since they traverse the same path except in opposite directions. Hence
Z
XZ
(D.10)
g dy = g dy,
j
j
and summing over j yields (D.2). Meanwhile, one can typically divide into domains of
type II, as depicted in Fig. D.5, get (D.8) on each such domain, and sum over j to get
(D.3).
260
(D.12)
ZZ
Z
u
u
+i
dx dy = i u (dx + i dy)
x
y
Z
= i u dz.
(D.13)
ZZ
Z
u
u
i
dx dy = i u (dx i dy)
x
y
Z
= i u dz.
In case u is holomorphic and u0 (z) = v(z), the integrand on the left side of (D.13) is 2v(z).
Such an identity (together with a further limiting argument) is useful in Exercise 7 of 30.
261
E. Poincar
e metrics
Recall from 22 that the upper half plane
(E.1)
U = {z C : Im z > 0}
az + b
a b
(E.2)
LA (z) =
, A=
,
c d
cz + d
of the form
(E.3)
D = {z C : z < 1}
a b
(E.5)
B SU (1, 1), i.e., B =
, a2 b2 = 1.
b a
These groups are related by the holomorphic diffeomorphism
zi
1 i
(E.6)
LA0 = : U D, A0 =
, i.e., (z) =
.
1 i
z+i
Here we produce metric tensors on U and D that are invariant under these respective group
actions. These are called Poincare metrics.
We start with U, and the metric tensor
(E.7)
gjk (z) =
1
jk ,
y2
dsU =
1
dz.
y
(z) = z + , R,
262
Note that
(E.10)
= LT , r = LDr ,
T =
1
0 1
, Dr =
r1/2
0
0
r1/2
The group generated by these operations is not all of Sl(2, R). We proceed as follows.
We pull the metric tensor (E.7) back to the disk D, via = 1 , given by
(E.11)
(z) =
1z+1
,
i z1
: D U.
We have
(E.12)
dz = 0 (z) dz =
2i
dz,
(z 1)2
and hence
1
 dz
Im (z)
2i
 0 (z) dz
=
(z) (z)
4
dz
= z+1 z+1
(z 1)(z 1)
z1 + z1
2
=
dz.
1 z2
dsU =
(E.13)
dsD =
2
dz,
1 z2
or
(E.15)
hjk (z) =
4
jk .
(1 z2 )2
LTe , LD
er ,
1
e
Te = A0 T A1
0 , Dr = A0 Dr A0 .
(z) = ei z,
R.
263
Here
(E.18)
= LR ,
R =
ei/2
0
ei/2
(E.19)
when
(E.20)
a,b (z) =
az + b
,
bz + a
a2 b2 = 1.
In fact,
(E.21)
dz
,
(bz + a)2
and hence
2
0 (z) dz
1 a,b (z)2 a,b
2 dz
=
bz + a2 az + b2
2 dz
=
1 z2
= dsD .
a,b dsD =
(E.22)
: D.
The pull back of the Poincare metric on D via is called the Poincare metric on . Note
e : D is another holomorphic diffeomorphism, then
e 1 : D D is a
that if
264
: D .
or
(E.26)
ds = A (z) dz,
AD (z) =
2
.
1 z2
The following is a definitive result on the class of domains to which such a construction
applies.
Theorem E.2. If C is a connected open set and C \ contains at least two points,
then there is a holomorphic covering map (E.24).
This is part of the celebrated Uniformization Theorem, of which one can read a careful
account in [For]. We will not give a proof of Theorem E.2 here. A proof using basic results
about partial differential equations is given in [MaT]. We recall that 25 establishes this
result for = C \ {0, 1}. To see how Theorem E.2 works for D = D \ {0}, note that
(E.29)
: U D ,
(z) = eiz ,
is a holomorphic covering map, and composing with the inverse of in (E.6) yields a
holomorphic covering map D D .
The following interesting result is a geometric version of the Schwarz lemma.
265
Proposition E.3. Assume O and are domains in C with Poincare metrics, inherited
from holomorphic coverings by D, and F : O is holomorphic. Then F is distancedecreasing.
What is meant by distance decreasing is the following. Let : [a, b] O be a smooth
path. Its length, with respect to the Poincare metric on O, is
Z
(E.30)
`O () =
The assertion that F is distance decreasing is that, for all such paths ,
(E.31)
` (F ) `O ().
Rb
Note that ` (F ) = a A (F (s))(F )0 (s) ds and (F )(s) = F 0 ((s)) 0 (s), so
the validity of (E.31) for all paths is equivalent to
(E.32)
z O.
To prove Proposition E.3, note that under the hypotheses given there, F : O lifts
to a holomorphic map G : D D, and it suffices to show that any such G is distance
decreasing, for the Poincare metric on D, i.e.,
(E.33)
z0 D.
Now, given z0 D, we can compose G on the right with a linear fractional transformation
of the form (E.20), taking 0 to z0 , and on the left by such a linear fractional transformation,
taking G(z0 ) to 0, obtaining
(E.34)
H : D D holomorphic, H(0) = 0,
H 0 (0) 1,
H(z) z.
266
pj = f (aj ) 0,
qj = f (bj ) 1,
`D (j ) 0,
`D (j ) 0.
j = f j ,
j = f j
`C (
j ) 0,
`C (
j ) 0.
`C (j ) = `D (
j ).
Now j contains points that tend to D (in the Euclidean metric) as j , so the fact
that `D (
j ) follows easily from the formula (E.14).
Returning to the setting of (E.37)(E.40), we deduce that
Given > 0, N < such than whenever j N
(E.42)
If, e.g., < 1/4, this cannot hold for a function f that is holomorphic on a neighborhood
of the annular region bounded by j and j . (Cf. 6, Exercise 2.) Thus our assumption
on f contradicts the CasoratiWeierstrass theorem, and Proposition E.4 is proven.
267
p(z) = an z n + + a1 z + a0
= an z n 1 + O(z 1 ) , z ,
which implies
(F.2)
lim p(z) = .
z
zR
we deduce that
(F.4)
zR
zC
p(z0 ) = a 6= 0.
We can write
(F.7)
p(z0 + ) = a + q(),
268
q() = b k + O( k+1 ),
0,
so, uniformly on S 1 = { :  = 1}
(F.9)
p(z0 + ) = a + b k k + O(k+1 ),
& 0.
b k
a
= ,
b
a
p(z0 + ) = a 1 k + O(k+1 ),
a
which contradicts (F.5) for > 0 small enough. Thus (F.6) is impossible. This proves
Lemma F.2, hence Theorem F.1.
269
then f (x) f (x) as 0, for each x R. In fact it folows easily that f f uniformly
on R, whenever f C0 (R). In particular, given > 0, there exists > 0 such that
(G.3)
x R.
Now note that, for each > 0, f (x) extends from x R to the entire holomorphic
function
Z
2
1
e(zy) /4 f (y) dy.
(G.4)
F (z) =
4
That this integral is absolutely convergent for each z C is elementary, and that it is
holomorphic in z can be deduced from Moreras theorem. It follows that F (z) has a
power series expansion,
X
(G.5)
F (z) =
an ()z n ,
n=0
where
(G.7)
PN, (z) =
N
X
an ()z n .
n=0
x [a, b].
270
f, g V = f + g V,
a C, f V = af V,
f + g = g + f, (f + g) + h = f + (g + h), f + 0 = f, f + (f ) = 0.
a(bf ) = (ab)f,
1 f = f.
a(f + g) = af + ag,
(a + b)f = af + bf.
These properties are readily verified for the function spaces arising in 1314.
An inner product on a complex vector space V assigns to elements f, g V the quantity
(f, g) C, in a fashion that obeys the following three rules:
(a1 f1 + a2 f2 , g) = a1 (f1 , g) + a2 (f2 , g),
(H.6)
(f, g) = (g, f ),
(f, f ) > 0
unless f = 0.
A vector space equipped with an inner product is called an inner product space. For
example,
Z
1
(H.7)
(f, g) =
f ()g() d
2
S1
271
(H.9)
ak 2 < .
k=
(ak ), (bk ) =
ak bk .
k=
kf k =
(f, f )
is the norm on V associated with the inner product. Generally, a norm on V is a function
f 7 kf k satisfying
(H.12)
(H.13)
(H.14)
kaf k = a kf k,
kf k > 0
a C, f V,
unless f = 0,
kf + gk kf k + kgk.
The property (H.14) is called the triangle inequality. A vector space equipped with a norm
is called a normed vector space. We can define a distance function on such a space by
(H.15)
d(f, g) = kf gk.
while
(H.17)
Thus to establish (H.17) it suffices to prove the following, known as Cauchys inequality.
272
Proposition H.1. For any inner product on a vector space V , with kf k defined by (H.11),
(H.18)
f, g V.
which implies
(H.20)
2 Re(f, g) kf k2 + kgk2 ,
f, g V.
f, g V.
f, g V, t (0, ).
If we take t2 = kgk/kf k, we obtain the desired inequality (H.18). This assumes f and g
are both nonzero, but (H.18) is trivial if f or g is 0.
An inner product space V is called a Hilbert space if it is a complete metric space, i.e.,
if every Cauchy sequence (f ) in V has a limit in V . The space `2 has this completeness
property, but C(S 1 ), with inner product (H.7), does not. Appendix A describes a process
of constructing the completion of a metric space. When appied to an incomplete inner
product space, it produces a Hilbert space. When this process is applied to C(S 1 ), the
completion is the space L2 (S 1 ), briefly discussed in 13. This result is essentially the
content of Propositions A and B, stated in 13.
There is a great deal more to be said about Hilbert space theory, but further material
is not needed here. One can consult a book on functional analysis, or the appendix on
functional analysis in Vol. 1 of [T2].
273
I. 2 is Irrational
The following proof that 2 is irrational follows a classic argument of D. Niven. The
idea is to consider
Z
1
(I.1)
In =
n (x) sin x dx, n (x) = xn ( x)n .
n!
0
Clearly In > 0 for each n N, and In 0 very fast, faster than geometrically. The next
key fact is that In is a polynomial of degree n in 2 with integer coefficients:
(I.2)
In =
n
X
cnk 2k ,
cnk Z.
k=0
n
X
k=0
But the left side of (I.3) is an integer for each n, while by the estimate on (I.1) mentioned
above the right side belongs to the interval (0, 1) for large n, yielding a contradiction. It
remains to establish (I.2).
A method of computing the integral in (I.1), which works for any polynomial n (x)) is
the following. One looks for an antiderivative of the form
(I.4)
hence
(I.6)
One can exploit the nilpotence of x2 on the space of polynomials of degree 2n and set
Fn (x) = (I + x2 )1 n (x)
(I.7)
n
X
k=0
(1)k (2k)
n (x).
274
Then
d 0
Fn (x) sin x Fn (x) cos x = n (x) sin x.
dx
(I.8)
(I.9)
the last identity holding for n (x) as in (I.1) because then n ( x) = n (x) and hence
Fn ( x) = Fn (x). For the first identity in (I.9), we use the defining property that
sin = 0 while cos = 1.
(2k)
In light of (I.7), to prove (I.2) it suffices to establish an analogous property for n (0).
Comparing the binomial formula and Taylors formula for n (x):
n
1 X
` n
n` xn+` ,
(1)
n (x) =
n!
`
and
`=0
(I.10)
n (x) =
2n
X
k=0
1 (k)
n (0)xk ,
k!
we see that
(I.11)
k =n+`
(k)
n (0)
` (n
+ `)! n n`
,
n!
`
n (n
+ `)! n 2(k`)
.
n!
`
= (1)
so
(I.12)
2k = n + `
(2k)
Of course n
In fact,
(I.13)
(2k)
n (0)
= (1)
n+`
=
n(n 1) (n ` + 1).
n
275
J. Eulers constant
Here we say more about Eulers constant, introduced in (18.17), in the course of producing the Euler product expansion for 1/(z). The definition
(J.1)
n
X
1
= lim
log(n + 1)
n
k
k=1
(J.2)
(z) =
n=1 (1
+ z/n)ez/n implies
0 (1) = .
(J.3)
Thus we have the integral formula
Z
(J.4)
(log t)e
dt
(log t)et dt
= a b .
We can apply integration by parts to both the integrals in (5), using et = (d/dt)(et 1)
on the first and et = (d/dt)et on the second, to obtain
Z
(J.6)
a =
0
1 et
dt,
t
b =
1
et
dt.
t
Using the power series for et and integrating term by term produces a rapidly convergent series for a :
(J.7)
a =
X
(1)k1
k=1
k k!
276
Before producing infinite series representations for b , we note that the change of variable
t = sm gives
Z sm
e
(J.8)
b = m
ds,
s
1
which is very well approximated by the integral over s [1, 10) if m = 2, for example.
To produce infinite series for b , we can break up [1, ) into intervals [k, k + 1) and
take t = s + k, to write
(J.9)
b =
X
ek
k=1
Z
k ,
k =
0
et
dt.
1 + t/k
Note that 0 < k < 1 1/e for all k. For k 2 we can write
X
1 j
k =
j ,
k
j=0
(J.10)
j =
tj et dt.
Ej (t) =
j
X
t`
`=0
`!
Then
(J.12)
hence
(J.13)
d
tj
Ej (t)et = Ej1 (t) Ej (t) et = et ,
dt
j!
so
Z
tj et dt = j!Ej (t)et + C.
(J.14)
In particular,
Z
j =
j t
t e
0
(J.15)
1 X 1
dt = j! 1
e
`!
j
`=0
j!
1
e
`!
`=j+1
1 1
1
=
+
+ .
e j + 1 (j + 1)(j + 2)
=
277
1 =
1
(J.16)
et
dt
t
X
(1)j
j=0
j!
= log 2 +
tj1 dt
X
(1)j
j=1
j j!
(2j 1).
X
X
ek 1 j
(1)j j
b =
j + log 2 +
(2 1),
k
k
j j!
j=0
j=1
k=2
with j given by (J.15). We can reverse the order of summation of the double series and
write
(J.18)
b =
(1) j j + log 2 +
j=0
X
(1)j
j=1
j j!
(2j 1).
with
X
ek
j =
.
k j+1
(J.19)
k=2
Note that
(J.20)
(j+1)
0 < j < 2
ek < 2(j+3) ,
k=2
while (J.15) readily yields 0 < j < 1/ej. So one can expect 15 digits of accuracy by
summing the first series in (J.18) over 0 j 50 and the second series over 0 j 32,
assuming the ingredients j and j are evaluated sufficiently accurately. It suffices to sum
(J.19) over 2 k 40 2j/3 to evaluate j to sufficient accuracy.
Note that the quantities j do not have to be evaluated independently. Say you are
summing the first series in (J.18) over 0 j 50. First evaluate 50 using 20 terms in
(J.15), and then evaluate inductively 49 , . . . , 0 using the identity
(J.21)
j1 =
j
1
+
,
je
j
R1
0
tj et dt.
278
If we sum the series (J.7) for a over 1 k 20 and either sum the series (J.18) as
described above or have Mathematica numerically integrate (J.8), with m = 2, to high
precision, we obtain
(J.22)
0.577215664901533,
X
n=1
n ,
1
n =
n
n+1
n
1
1
dx
= log 1 +
.
x
n
n
n
1
1o
n = (x, y) : n x n + 1,
y
.
x
n
This region contains the triangle Tn with vertices (n, 1/n), (n+1, 1/n), and (n+1, 1/(n+1)).
The region n \ Tn is a little sliver. Note that
(J.25)
Area Tn = n =
1 1
1
,
2 n n+1
and hence
(J.26)
n =
n=1
1
.
2
Thus
(J.27)
1
= (1 1 ) + (2 2 ) + (3 3 ) + .
2
Now
(J.28)
1 1 =
3
log 2,
4
(J.29)
1
1
1
3 + 4
2
2n
3n
4n
1
1
1
n = 2 3 + 4 ,
2n
2n
2n
n =
279
n =
1
1
1
= 2
2n(n + 1)
2n 1 +
1
n
1 1 X 1
1 1 X 1
1
= (1 1 ) +
+ ,
2
2 3
n3
2 4
n4
n2
n2
or, with
(k) =
(J.32)
X 1
,
nk
n1
we have
(J.33)
1 1
1 1
1 3
=
log 2 +
[(3) 1]
[(4) 1] + ,
2
4
2 3
2 4
(J.34)
1
[(3) 1] 0.0336762,
6
1
[(4) 1] 0.0205808,
4
3
[(5) 1] 0.0110783.
10
The estimate
(J.35)
Z
X 1
k
<2 +
xk dx
nk
2
n2
implies
(J.36)
1
0<
[(k) 1] < 2k ,
2 k
280
X
1
1
1
(z; ) = 2 +
2 .
z
(z )2
06=
This converges rather slowly, so another method must be used to evaluate (z; ) rapidly.
The classical method, which we describe below, involves a representation of in terms of
theta functions. It is most conveniently described in case
(K.2)
generated by 1 and ,
Im > 0.
To pass from this to the general case, we can use the identity
(K.3)
(z; a) =
1 z
; .
a2
a
The material below is basically a summary of material from 32, assembled here to clarify
the important application to the task of the rapid evaluation of (K.1).
To evaluate (z; ), which we henceforth denote (z), we use the following identity:
(K.4)
0 (0) (z) 2
2
1
(z) = e1 +
.
2 (0) 1 (z)
See (32.20). Here e1 = (1 /2) = (1/2), and the theta functions j (z) (which also
depend on ) are defined as follows (cf. (32.6)(32.10)):
1 (z) = i
2 (z) =
(K.5)
3 (z) =
4 (z) =
n=
p2n1 q (n1/2) ,
n=
X
n=
p2n q n ,
2
(1)n p2n q n .
n=
Here
(K.6)
p = eiz ,
q = ei ,
281
with as in (K.2).
The functions 1 and 2 appear in (K.4). Also 3 and 4 arise to yield a rapid evaluation
of e1 (cf. (32.33)):
(K.7)
e1 =
2
3 (0)4 + 4 (0)4 .
3
01 (0) =
(K.8)
n=
n=1
n=1
also formulas for 01 (0) and j (0), which appear in (K.4) and (K.7), can be rewritten:
01 (0)
= 2
2 (0) = 2
(K.10)
n=1
X
(n1/2)2
n=1
3 (0) = 1 + 2
4 (0) = 1 + 2
qn ,
n=1
(1)n q n .
n=1
Rectangular lattices
We specialize to the case where is a rectangular lattice, of sides 1 and L, more
precisely:
(K.11)
L > 0.
q = eL .
282
Since q is real, we see that the quantities 01 (0) and j (0) in (K.10) are real. It is also
convenient to calculate the real and imaginary parts of j (z) in this case. Say
(K.13)
z = u + iv,
(K.14)
We then have
Re(i1 (z)) =
2
(1)n q (n1/2) e(2n1)v e(2n1)v cos(2n 1)u,
n=1
(K.15)
Im(i1 (z)) =
2
(1)n q (n1/2) e(2n1)v + e(2n1)v sin(2n 1)u,
n=1
and
Re 2 (z) =
(K.16)
2
q (n1/2) e(2n1)v + e(2n1)v cos(2n 1)u,
n=1
Im 2 (z) =
2
q (n1/2) e(2n1)v e(2n1)v sin(2n 1)u.
n=1
We can calculate these quantities accurately by summing over a small range. Let us
insist that
(K.17)
1
1
u< ,
2
2
L
L
v< ,
2
2
and assume
(K.18)
L 1.
Then
(K.19)
(n1/2)2 (2n1)v
q
e(n2 3n+5/4)L ,
e
and since
e <
(K.20)
1
,
20
for n = 5,
25
for n = 6,
< 2 10
283
with rapid decrease for n > 6. Thus, summing over 1 n 5 will give adequate approximations.
For z = u + iv very near 0, where 1 vanishes and has a pole, the identity
(0) (z) 2
1
2
1
=
,
(z) e1
01 (0) 2 (z)
(K.22)
(K.23)
k = 2n 1 {1, 3, 5, 7, 9},
when v is very small, since then (K.23) is the difference between two quantities close to 1,
so evaluating ekv and ekv separately and subtracting can lead to an undesirable loss of
accuracy. In case k = 1, one can effect this cancellation at the power series level and write
(K.24)
=2
j1,odd
(v)j
.
j!
If v 102 , summing over j 7 yields substantial accuracy. (If v > 102 , separate
evaluation of ekv and ekv should not be a problem.) For other values of k in (K.23),
one can derive from
(xk 1) = (x 1)(xk1 + + 1)
(K.25)
the identity
(K.26)
kv
kv
= (e
k1
X
e(2`(k1))v ,
`=0
which in concert with (K.24) yields an accurate evaluation of each term in (K.23).
Remark. If (K.11) holds with 0 < L < 1, one can use (K.3), with a = iL, to transfer to
the case of a lattice generated by 1 and i/L.
284
References
[Ahl]
[BN]
[BB]
[CG]
[Cl]
[Con]
[Ed]
[FK]
[For]
[Gr]
[Hil]
[Kra]
[Law]
[MaT]
[MM]
[Mil]
[Mun]
[Sch]
[Sp]
[T0]
[T]
[T2]
[T3]
[Ts]
285
Index
Abel inversion problem, 217
analytic continuation, 85
argument principle, 135
arc length, 43
ArzelaAscoli theorem, 160, 240
Arithmeticgeometric mean, 223
Bernoulli numbers, 97, 201
binomial coefficient, 37, 40, 134
branched covering, 227
CasoratiWeierstrass theorem, 90
Cauchy integral formula, 56
Cauchy integral theorem, 56
CauchyRiemann equation, 24
Cauchys inequalities, 62, 108
Cauchy sequence 10, 233
chain rule, 20, 26, 245
compact space, 233
complex differentiable, 20
conformal map, 162
connected, 242
contraction mapping theorem, 242
convolution, 111, 120, 128
cos 43
covering map, 158, 175
derivative, 13, 20, 245
Dirichlet problem, 109, 193
domain
elliptic function, 199, 205
elliptic integral, 206, 217
of first, second, third kind, 220
essential singularity, 90
Eulers constant, 143, 275
Eulers formula, 43
Euler product formula, 143
exponential function, 15, 21, 39
286
287
Much more than documents.
Discover everything Scribd has to offer, including books and audiobooks from major publishers.
Cancel anytime.